feat: PraisonAIUI host integration (Patterns B/C) + core protocol bridges#1706
Conversation
…col bridges. Introduces configure_host/build_host_app, optional L1 bridges (hooks, workflows, usage, approvals, schedules), refactors bundled default apps, and adds session/telemetry protocols plus integration tests so aiui can run standalone or wired via set_backend injection.
Qodo reviews are paused for this user.Troubleshooting steps vary by plan Learn more → On a Teams plan? Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center? |
📝 WalkthroughWalkthroughAdds a unified PraisonAIUI host bootstrap (Patterns B/C), public token-collector accessor and usage-query adapters, session metadata hoisting/persistence, integration bridges, refactored UI apps using configure_host/is_legacy_host, tool-call end stream events, and supporting tests and docs. ChangesHost App Bootstrap & Token Telemetry
Estimated code review effort🎯 4 (Complex) | ⏱️ ~75 minutes Possibly related PRs
Suggested labels
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
@copilot Do a thorough review of this PR. Read ALL existing reviewer comments above from Qodo, Coderabbit, and Gemini first — incorporate their findings. Review areas:
|
|
@claude You are the FINAL architecture reviewer. If the branch is under MervinPraison/PraisonAI (not a fork), you are able to make modifications to this branch and push directly. SCOPE: Focus ONLY on Python packages (praisonaiagents, praisonai). Do NOT modify praisonai-rust or praisonai-ts. Read ALL comments above from Gemini, Qodo, CodeRabbit, and Copilot carefully before responding. Phase 1: Review per AGENTS.md
Phase 2: FIX Valid Issues Phase 3: Final Verdict |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a robust integration layer between PraisonAI and PraisonAIUI, implementing Pattern B and Pattern C host architectures. It refactors default applications to use a new centralized configure_host bootstrap, enhances session metadata persistence, and adds bridge protocols for hooks, workflows, and token usage tracking. Feedback identifies a fragile implementation using name mangling for stream emitters and suggests a more efficient, single-pass approach for calculating token usage summaries.
| agent_id=self.name, | ||
| metadata={"duration_ms": _duration_ms}, | ||
| )) | ||
| getattr(self, "_Agent__stream_emitter", None).emit(StreamEvent( |
There was a problem hiding this comment.
Accessing the private attribute _Agent__stream_emitter via getattr with a hardcoded mangled name is fragile. This implementation depends on the mixin being applied to a class named exactly Agent and will break if the attribute is renamed or if the class hierarchy changes. It is recommended to use a protected attribute (e.g., _stream_emitter) or provide a public/protected method in the Agent class to access the emitter.
| def get_summary(self) -> Dict[str, Any]: | ||
| records = self._sink.records | ||
| total_in = sum(r.get("input_tokens", 0) for r in records) | ||
| total_out = sum(r.get("output_tokens", 0) for r in records) | ||
| by_model: Dict[str, Dict[str, int]] = {} | ||
| by_agent: Dict[str, Dict[str, int]] = {} | ||
| for r in records: | ||
| model = r.get("model", "unknown") | ||
| agent = r.get("agent_name", "unknown") | ||
| by_model.setdefault(model, {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0}) | ||
| by_agent.setdefault(agent, {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0}) | ||
| for bucket, key in ((by_model, model), (by_agent, agent)): | ||
| bucket[key]["input_tokens"] += r.get("input_tokens", 0) | ||
| bucket[key]["output_tokens"] += r.get("output_tokens", 0) | ||
| bucket[key]["total_tokens"] += r.get("total_tokens", 0) | ||
| return { | ||
| "total_requests": len(records), | ||
| "total_input_tokens": total_in, | ||
| "total_output_tokens": total_out, | ||
| "total_tokens": total_in + total_out, | ||
| "by_model": by_model, | ||
| "by_agent": by_agent, | ||
| } |
There was a problem hiding this comment.
The get_summary method is inefficient as it performs three separate passes over the records list (two sum() calls and one for loop). Additionally, the total_tokens in the returned dictionary is calculated as total_in + total_out, which ignores other token types (like cached or reasoning tokens) that might be present in the records. This makes the overall total inconsistent with the breakdown in by_model and by_agent which use r.get("total_tokens", 0). A single-pass approach would be more efficient and consistent.
def get_summary(self) -> Dict[str, Any]:
records = self._sink.records
total_in = 0
total_out = 0
total_all = 0
by_model: Dict[str, Dict[str, int]] = {}
by_agent: Dict[str, Dict[str, int]] = {}
for r in records:
r_in = r.get("input_tokens", 0)
r_out = r.get("output_tokens", 0)
r_total = r.get("total_tokens", r_in + r_out)
total_in += r_in
total_out += r_out
total_all += r_total
model = r.get("model", "unknown")
agent = r.get("agent_name", "unknown")
for bucket, key in ((by_model, model), (by_agent, agent)):
stats = bucket.setdefault(key, {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0})
stats["input_tokens"] += r_in
stats["output_tokens"] += r_out
stats["total_tokens"] += r_total
return {
"total_requests": len(records),
"total_input_tokens": total_in,
"total_output_tokens": total_out,
"total_tokens": total_all,
"by_model": by_model,
"by_agent": by_agent,
}
|
Claude finished @MervinPraison's task —— View job Final Architecture Review ✅Phase 1: Architecture Review per AGENTS.md
Phase 2: Fix Valid Issues from Previous Reviewers
Phase 3: Final Verdict
✅ Architecture Review SummaryAGENTS.md Compliance: Perfect ✅
Backward Compatibility: Perfect ✅
Performance Impact: Clean ✅
🔧 Issues Fixed1. Fragile Name Mangling (
|
Greptile SummaryThis PR adds a
Confidence Score: 4/5Safe to merge with one small fix: the list_recent(limit=0) bug in InMemoryUsageQuery returns the full usage history instead of an empty slice, which could leak data to pagination callers. The integration layer is well-structured and bridge failures are all swallowed gracefully. The one concrete defect is in InMemoryUsageQuery.list_recent: Python's -0 == 0 causes records[-0:] to return all records when limit=0 is passed, violating the contract and potentially exposing the entire token-usage history to consumers expecting an empty result. src/praisonai-agents/praisonaiagents/telemetry/protocols.py (InMemoryUsageQuery.list_recent) Important Files Changed
Sequence DiagramsequenceDiagram
participant App as default_app.py (import)
participant CH as configure_host()
participant AUI as praisonaiui
participant Prov as PraisonAIProvider
participant Bridges as setup_bridges()
participant WSGI as create_host_app()
App->>CH: configure_host(pages, agent_kwargs)
CH->>AUI: set_datastore(PraisonAISessionDataStore())
CH->>AUI: set_style / set_branding / set_pages / set_dashboard
alt not legacy mode
CH->>Prov: "PraisonAIProvider(**agent_kwargs)"
CH->>AUI: set_provider(provider)
CH->>Bridges: setup_bridges()
Bridges->>AUI: backends.set_backend(hooks, workflows, usage)
end
CH-->>App: "_CONFIGURED = True"
App->>WSGI: create_host_app()
WSGI->>AUI: create_app()
AUI-->>App: Starlette app
Reviews (2): Last reviewed commit: "fix: address gemini review findings in t..." | Re-trigger Greptile |
…y access - Replace getattr(self, '_Agent__stream_emitter', None) with self.stream_emitter - Optimize token summary calculation with single-pass algorithm - Fix inconsistent total_tokens calculation in telemetry protocols Co-authored-by: Mervin Praison <MervinPraison@users.noreply.github.com>
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
| elif name == 'build_host_app': | ||
| from .integration.host_app import build_host_app | ||
| return build_host_app | ||
| elif name == 'configure_host': | ||
| from .integration.host_app import configure_host | ||
| return configure_host | ||
| elif name == 'AgentApp': |
There was a problem hiding this comment.
AgentOS entry was removed from __getattr__ and replaced with the two new integration exports. Any code that does from praisonai import AgentOS will now raise AttributeError: module 'praisonai' has no attribute 'AgentOS' even though AgentOS is still importable from praisonai.app. The AgentApp alias block still works because it has its own branch, but the primary name is silently broken.
| elif name == 'build_host_app': | |
| from .integration.host_app import build_host_app | |
| return build_host_app | |
| elif name == 'configure_host': | |
| from .integration.host_app import configure_host | |
| return configure_host | |
| elif name == 'AgentApp': | |
| elif name == 'AgentOS': | |
| from .app import AgentOS | |
| return AgentOS | |
| elif name == 'build_host_app': | |
| from .integration.host_app import build_host_app | |
| return build_host_app | |
| elif name == 'configure_host': | |
| from .integration.host_app import configure_host | |
| return configure_host | |
| elif name == 'AgentApp': |
| def register_legacy_reply(aiui, *, agent_name: str = "PraisonAI"): | ||
| """Register callback-only reply handler when ``PRAISONAI_HOST_LEGACY=1``.""" | ||
|
|
||
| @aiui.reply | ||
| async def legacy_reply(message: str, session_id: str = "default"): | ||
| from praisonaiagents import Agent | ||
|
|
||
| agent = Agent( | ||
| name=agent_name, | ||
| instructions="You are a helpful assistant.", | ||
| llm="gpt-4o-mini", | ||
| ) | ||
| return agent.run(message) |
There was a problem hiding this comment.
Agent on every invocation, so each message starts with an empty context — no conversation history is preserved between turns. The previous per-default-app implementations all used a session_id-keyed cache to maintain continuity. This centralised version loses that behaviour for every app in legacy mode.
| def register_legacy_reply(aiui, *, agent_name: str = "PraisonAI"): | |
| """Register callback-only reply handler when ``PRAISONAI_HOST_LEGACY=1``.""" | |
| @aiui.reply | |
| async def legacy_reply(message: str, session_id: str = "default"): | |
| from praisonaiagents import Agent | |
| agent = Agent( | |
| name=agent_name, | |
| instructions="You are a helpful assistant.", | |
| llm="gpt-4o-mini", | |
| ) | |
| return agent.run(message) | |
| def register_legacy_reply(aiui, *, agent_name: str = "PraisonAI"): | |
| """Register callback-only reply handler when ``PRAISONAI_HOST_LEGACY=1``.""" | |
| _cache: dict = {} | |
| @aiui.reply | |
| async def legacy_reply(message: str, session_id: str = "default"): | |
| from praisonaiagents import Agent | |
| if session_id not in _cache: | |
| _cache[session_id] = Agent( | |
| name=agent_name, | |
| instructions="You are a helpful assistant.", | |
| llm="gpt-4o-mini", | |
| ) | |
| return await _cache[session_id].achat(message) |
| ) -> None: | ||
| """Apply PraisonAIUI host settings and wire L1 backends (unless legacy mode).""" | ||
| global _CONFIGURED | ||
|
|
||
| import praisonaiui as aiui | ||
| from praisonai.ui._aiui_datastore import PraisonAISessionDataStore | ||
|
|
||
| aiui.set_datastore(PraisonAISessionDataStore()) | ||
| aiui.set_style("dashboard") | ||
| aiui.set_branding(title=title, logo=logo) | ||
|
|
||
| if pages is not None: | ||
| aiui.set_pages(list(pages)) |
There was a problem hiding this comment.
configure_host has no idempotency guard
_CONFIGURED is only written at the end of this function and is never read inside it, so every call re-runs the full body: a new PraisonAISessionDataStore() is set, a new PraisonAIProvider is constructed and registered, and setup_bridges() is called again. Adding an early-return guard — if _CONFIGURED: return — at the top of the function would make it safe to call multiple times.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 18
Note
Due to the large number of review comments, Critical, Major severity comments were prioritized as inline comments.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (3)
src/praisonai-agents/praisonaiagents/session/README.md (1)
165-165:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winFix inconsistent session directory path documentation.
The comment on Line 165 states
Default: ~/.praison/sessions/but Line 35 documents the path as~/.praisonai/sessions/{session_id}.json. These paths are inconsistent and will confuse users about the actual storage location.📝 Proposed fix
- session_dir: Optional[str] = None, # Default: ~/.praison/sessions/ + session_dir: Optional[str] = None, # Default: ~/.praisonai/sessions/🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai-agents/praisonaiagents/session/README.md` at line 165, Update the inconsistent session directory path in the README: change the comment on the session_dir parameter (session_dir: Optional[str] = None, # Default: ~/.praison/sessions/) to match the documented storage location used elsewhere (e.g., ~/.praisonai/sessions/{session_id}.json). Ensure the text referencing the session_dir parameter and any examples use the exact same path string as the earlier documentation (the line that mentions ~/.praisonai/sessions/{session_id}.json) so both places consistently refer to ~/.praisonai/sessions/{session_id}.json.src/praisonai/praisonai/__init__.py (1)
133-142:⚠️ Potential issue | 🟠 Major | ⚡ Quick winRestore direct
AgentOSlazy resolution to avoid public API break.
__all__still exportsAgentOS(Line 14), but__getattr__no longer resolvesname == "AgentOS". This breaksfrom praisonai import AgentOSwithAttributeError.Proposed fix
elif name == 'configure_host': from .integration.host_app import configure_host return configure_host + elif name == 'AgentOS': + from .app import AgentOS + return AgentOS elif name == 'AgentApp': # Silent alias for AgentOS (backward compatibility) from .app import AgentOS return AgentOS🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai/praisonai/__init__.py` around lines 133 - 142, The module removed lazy resolution for AgentOS causing "from praisonai import AgentOS" to fail; restore handling in __getattr__ by adding a branch that returns the AgentOS symbol (importing AgentOS from .app) when name == "AgentOS" so the exported name in __all__ is resolved lazily; update the __getattr__ function to include this case alongside existing branches like 'build_host_app' and 'configure_host' to preserve backward compatibility.src/praisonai-agents/praisonaiagents/telemetry/protocols.py (1)
124-195: 🛠️ Refactor suggestion | 🟠 Major | 🏗️ Heavy liftKeep
protocols.pyinterface-only; move adapter implementations out.
InMemoryUsageQueryandTokenCollectorUsageQueryare concrete HOW implementations insideprotocols.py. They should live in dedicated adapter modules, with only Protocol contracts kept here.As per coding guidelines,
All major modules in the core SDK must have a protocols.py file defining WHAT (interface contracts), with separate adapter files implementing HOW (concrete implementations).🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai-agents/praisonaiagents/telemetry/protocols.py` around lines 124 - 195, protocols.py currently contains concrete adapter implementations (InMemoryUsageQuery, TokenCollectorUsageQuery) instead of only interface contracts; move the adapter classes out of protocols.py into new adapter modules (e.g., telemetry/adapters/in_memory_usage_query.py and telemetry/adapters/token_collector_usage_query.py) preserving their methods get_summary, list_recent, get_by_task, and get_by_agent, and leave protocols.py containing only the Protocol/ABC definitions for the usage query interface; update imports/exports so callers import the implementations from the new adapter modules and the interface from protocols.py.
🟡 Minor comments (4)
src/praisonai/tests/integration/test_aiui_host_agentic.py-22-31 (1)
22-31:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winUse a unique session ID per test run to avoid persistence collisions.
A fixed ID can pass due to stale sessions from prior runs.
Suggested fix
import os +from uuid import uuid4 @@ - session_id = "agentic-ac3-test" + session_id = f"agentic-ac3-{uuid4().hex}"🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai/tests/integration/test_aiui_host_agentic.py` around lines 22 - 31, The test uses a fixed session_id ("agentic-ac3-test") causing collisions; change the test to generate a unique session_id per run (e.g., using uuid.uuid4() or timestamp) before calling client.post("/run") so the created session is guaranteed unique; update references to session_id that follow (the client.post("/run") call, client.get("/sessions") check and the ids/assertion logic) to use the generated value so the assertion reliably verifies the newly created session.src/praisonai-agents/praisonaiagents/session/store.py-527-529 (1)
527-529:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winPreserve valid
0values in session summary fields.
ortreats0as falsy, sototal_tokens=0orcost=0can be replaced by fallback values (orNone) incorrectly.Proposed fix
- "total_tokens": data.get("total_tokens") or data.get("token_count") or (data.get("metadata") or {}).get("total_tokens"), - "cost": data.get("cost") or (data.get("metadata") or {}).get("cost"), + "total_tokens": ( + data["total_tokens"] if "total_tokens" in data + else data["token_count"] if "token_count" in data + else (data.get("metadata") or {}).get("total_tokens") + ), + "cost": ( + data["cost"] if "cost" in data + else (data.get("metadata") or {}).get("cost") + ),🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai-agents/praisonaiagents/session/store.py` around lines 527 - 529, The session summary construction incorrectly uses Python's `or` which treats 0 as falsy and will drop valid zero values for fields like `total_tokens` and `cost`; update the logic that sets `"model"`, `"total_tokens"`, and `"cost"` (the dict entries currently using `data.get("...") or ...`) to explicitly check for None (e.g., use conditional expressions or `if x is not None else` chains) so that 0 is preserved but None/falsy-missing values fall back to the other sources like `data.get("token_count")` or `(data.get("metadata") or {}).get("total_tokens")`.src/praisonai/praisonai/integration/host_app.py-51-52 (1)
51-52:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winHonor explicit empty
modulesconfiguration.Using a truthiness check here ignores an explicit empty list, so callers cannot intentionally clear modules. Use an explicit
Nonecheck instead.Suggested fix
- if modules: + if modules is not None: dashboard_opts["modules"] = list(modules)🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai/praisonai/integration/host_app.py` around lines 51 - 52, The current truthiness check "if modules:" prevents honoring an explicit empty list; change the conditional around the dashboard_opts assignment so it checks for None explicitly (e.g., use "if modules is not None:"), then set dashboard_opts["modules"] = list(modules) as before so callers can pass [] to clear modules while still ignoring a missing/None value.src/praisonai/praisonai/integration/bridges/approvals_bridge.py-26-29 (1)
26-29:⚠️ Potential issue | 🟡 Minor | ⚡ Quick winAvoid returning live mutable policy internals.
On Line 28, returning
_requirementsdirectly leaks mutable internal state to callers. Return a defensive copy.Proposed fix
registry = get_approval_registry() + requirements = getattr(registry, "_requirements", {}) + if isinstance(requirements, dict): + requirements = requirements.copy() return { "auto_approve_env": registry.is_env_auto_approve(), - "requirements": getattr(registry, "_requirements", {}), + "requirements": requirements, }🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai/praisonai/integration/bridges/approvals_bridge.py` around lines 26 - 29, The current return is leaking mutable internal state by returning registry._requirements directly; change the "requirements" value to a defensive copy (e.g., use copy.deepcopy(getattr(registry, "_requirements", {})) or at minimum dict(getattr(registry, "_requirements", {}))) so callers cannot mutate internals, and add the necessary import for copy if using deepcopy; keep the rest of the returned keys (e.g., "auto_approve_env" / registry.is_env_auto_approve()) unchanged.
🧹 Nitpick comments (4)
src/praisonai-agents/praisonaiagents/scheduler/README.md (1)
1-30: ⚡ Quick winConsider using Mintlify components for enhanced documentation.
The documentation currently uses plain Markdown. As per coding guidelines, documentation under
src/praisonai-agents/**/*.mdshould use Mintlify components such as CodeGroup, Steps, Tabs, Note/Warning callouts for better presentation. Consider restructuring with:
<Steps>for the "Use when" sections<CodeGroup>for the Python example<Note>callout to highlight the stateless vs. daemon distinctionAs per coding guidelines: "Documentation: Use Mintlify components (Accordion, Card, CodeGroup, Steps, Tabs, Note/Warning callouts, Frame, Mermaid diagrams)".
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai-agents/praisonaiagents/scheduler/README.md` around lines 1 - 30, Replace the plain Markdown in this README with Mintlify components: convert the "Use when" bullet lists into a <Steps> component for each of ScheduleRunner and ScheduleLoop, wrap the Python example in a <CodeGroup> block, and add a <Note> (or Warning) callout that highlights the stateless vs daemon distinction for ScheduleRunner and ScheduleLoop (mentioning ScheduleRunner, ScheduleLoop, FileScheduleStore, and ensure_schedule_runner to keep context). Keep the same headings and content but restructure into Mintlify components for improved presentation and consistency with the project's docs guidelines.src/praisonai-agents/praisonaiagents/session/README.md (1)
1-207: ⚖️ Poor tradeoffConsider using Mintlify components for enhanced documentation.
The documentation currently uses plain Markdown. As per coding guidelines, documentation under
src/praisonai-agents/**/*.mdshould use Mintlify components. Consider:
<CodeGroup>for the multiple Python examples<Accordion>for the "Advanced Usage" section<Note>callouts for important information (e.g., multi-process safety, context caching)<Card>components for the behavior matrix and metadata contract tableThis would improve readability and align with the project's documentation standards.
As per coding guidelines: "Documentation: Use Mintlify components (Accordion, Card, CodeGroup, Steps, Tabs, Note/Warning callouts, Frame, Mermaid diagrams)".
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai-agents/praisonaiagents/session/README.md` around lines 1 - 207, Convert the plain Markdown README sections into Mintlify components: wrap the multiple Python examples in the "Quick Start" and "Advanced Usage" sections using <CodeGroup>, turn the "Advanced Usage" block into an <Accordion> with separate panels (Direct Session Store Access, Custom Session Directory, Using with DB Adapter), mark Multi-Process Safety and Context Caching as <Note> callouts, and render the "Session metadata contract" and "Behavior Matrix" tables inside <Card> components; keep all API reference class names (DefaultSessionStore, SessionData, SessionMessage) and headings intact while replacing the corresponding Markdown code blocks and tables with the stated Mintlify components for consistent formatting.src/praisonai-agents/praisonaiagents/agent/tool_execution.py (1)
293-304: ⚡ Quick winConsider documenting the semantic difference between
TOOL_CALL_RESULTandTOOL_CALL_END.The new
TOOL_CALL_ENDevent is emitted immediately afterTOOL_CALL_RESULTwith an identical payload structure (except for the event type). While this appears intentional for AIUI integration, the code lacks inline documentation explaining why both events are needed and how consumers should interpret the distinction.Consider adding a comment above line 293 to clarify the semantic contract, for example:
# TOOL_CALL_END signals the tool call lifecycle is fully complete # (distinct from TOOL_CALL_RESULT which provides the result data). # Consumers can use END for UI state finalization, cleanup, or # confirming no additional events will be emitted for this call.This would help future maintainers and API consumers understand the event sequence contract.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai-agents/praisonaiagents/agent/tool_execution.py` around lines 293 - 304, Add an inline comment above the StreamEvent emission in tool_execution.py where _Agent__stream_emitter.emit(StreamEvent(..., type=StreamEventType.TOOL_CALL_END, ...)) is called to document the semantic difference between TOOL_CALL_RESULT and TOOL_CALL_END: state that TOOL_CALL_RESULT carries the tool's result payload while TOOL_CALL_END signals lifecycle completion (no further events for this tool_call_id/function_name) and is intended for UI finalization/cleanup; reference the StreamEvent/StreamEventType types, the emit call, and the tool_call_id to make the contract clear to future maintainers.src/praisonai-agents/praisonaiagents/agent/memory_mixin.py (1)
402-407: ⚡ Quick winPersist session stats only after assistant messages.
_persist_session_stats()currently runs for user turns too, which adds extra session-store writes without new token/cost data.Proposed fix
if role == "user": self._session_store.add_user_message(self._session_id, content) elif role == "assistant": self._session_store.add_assistant_message(self._session_id, content) - self._persist_session_stats() + self._persist_session_stats()🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai-agents/praisonaiagents/agent/memory_mixin.py` around lines 402 - 407, The code calls _persist_session_stats() for both user and assistant turns causing unnecessary writes; change the logic in the try block inside memory_mixin.py so that _persist_session_stats() is invoked only after adding an assistant message (i.e., move the _persist_session_stats() call into the elif role == "assistant" branch immediately after self._session_store.add_assistant_message(self._session_id, content)), leaving user handling (self._session_store.add_user_message) unchanged and preserving the surrounding exception handling.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@src/praisonai-agents/praisonaiagents/hooks/query_protocol.py`:
- Around line 8-13: The HooksQueryProtocol interface (class HooksQueryProtocol
with method list_hooks_for_api) should be moved out of the query_protocol module
into the module that defines WHAT contracts (hooks.protocols): create the
Protocol class there, preserve `@runtime_checkable` and the method signature, then
update any imports in code that currently import HooksQueryProtocol from
query_protocol to import it from hooks.protocols and remove the old definition
from query_protocol; ensure typing imports (Protocol, List, Dict) remain present
and adjust package exports if necessary.
In `@src/praisonai-agents/praisonaiagents/skills/protocols.py`:
- Around line 178-193: The function list_skills_for_api is a concrete
SkillManager-backed adapter and should be moved out of protocols.py; create a
new adapter module (e.g., skills_adapter or skill_catalog_adapter) that
implements list_skills_for_api by importing SkillManager and calling
manager.get_available_skills(), keep protocols.py limited to interface
definitions only (remove list_skills_for_api from protocols.py), update any
imports to reference the new adapter module, and ensure the adapter returns the
same dict shape ("name", "description", "location") as before.
- Around line 180-193: The current broad except Exception hides real
discovery/runtime errors; restrict the graceful fallback to import failures only
by isolating the "from .manager import SkillManager" in its own try/except
ImportError block (or ModuleNotFoundError) and return [] only on that import
error, then instantiate SkillManager and call manager.get_available_skills()
without swallowing exceptions so runtime/catalog errors surface; reference the
SkillManager import, the manager variable, and manager.get_available_skills()
when making this change.
In `@src/praisonai-agents/praisonaiagents/telemetry/protocols.py`:
- Around line 130-152: The get_summary method under InMemoryUsageQuery
incorrectly computes the top-level "total_tokens" as total_input_tokens +
total_output_tokens which can under-report when records include their own
total_tokens (e.g., cached tokens); change the aggregation to compute
total_tokens by summing r.get("total_tokens", 0) across records (and ensure any
per-bucket increments consistently use r.get("total_tokens", 0) as already done
in the bucket loop) so the returned "total_tokens" matches the summed record
totals and per-model/per-agent buckets.
In `@src/praisonai/praisonai/integration/_legacy_handlers.py`:
- Around line 10-18: The async function legacy_reply currently calls the
synchronous Agent.run which can block the event loop; update legacy_reply to
call and await the asynchronous method Agent.arun(message) instead (i.e.,
replace agent.run(message) with await agent.arun(message)) so the handler
remains non-blocking; ensure the created agent object (Agent(...)) supports arun
and propagate await back to the caller.
In `@src/praisonai/praisonai/integration/bridges/approvals_bridge.py`:
- Around line 16-17: The current broad except ImportError: pass around the
import of praisona iagents hides real ImportError/ModuleNotFound issues; update
both occurrences (the try/except blocks that import "praisonaiagents") to only
suppress the missing-optional-dependency error: catch ModuleNotFoundError or
catch ImportError as e and re-raise unless the missing module name is
"praisonaiagents" (i.e., only swallow the error when e.name == "praisonaiagents"
or the message indicates that module); for any other ImportError, re-raise so
real import failures aren't silently ignored. Ensure both occurrences in
approvals_bridge.py are updated.
In `@src/praisonai/praisonai/integration/bridges/workflows_service.py`:
- Around line 31-33: The YAML loading (load_workflow_yaml when yaml_path is
truthy) occurs outside the try/except so parse/file errors escape the
failure-response contract; move the yaml load inside the existing try block or
wrap it in its own try/except and on exception return the same run-record
failure shape (e.g. {"status":"failed", "message": str(e)} or delegate to the
existing error handling path) and only merge into config after successful load;
apply the same change to the other yaml-loading section that mirrors this logic
(the second yaml_path handling block) so all yaml/file/parse errors produce the
intended "status":"failed" run record.
In `@src/praisonai/praisonai/ui_agents/default_app.py`:
- Around line 125-130: The legacy Agent in default_app.py uses
os.getenv("MODEL_NAME") directly, causing it to ignore PRAISONAI_MODEL which
configure_host() prefers; update the Agent creation (the Agent(...) call when
populating _agents_cache for session_id) to resolve the model the same way as
configure_host()—prefer PRAISONAI_MODEL first, then MODEL_NAME, then the
existing default ("gpt-4o-mini")—so the llm argument matches host configuration.
- Around line 55-57: The current except block in
src/praisonai/praisonai/ui_agents/default_app.py swallows all exceptions (print
+ return []) causing YAML parse/import errors to silently fall back to
DEFAULT_AGENTS; change the error handling so only a missing file falls back:
specifically, catch FileNotFoundError (or verify os.path.exists) and return
DEFAULT_AGENTS in that branch, but for other exceptions raised while
loading/parsing agents.yaml (e.g., YAML parse or import errors in the try blocks
around the code that references agents.yaml) log the full error and re-raise or
propagate it instead of returning []; apply the same change to the other
try/except sections noted (the block covering lines ~60-75) so only file-missing
returns the default and all other failures surface.
In `@src/praisonai/praisonai/ui_bot/default_app.py`:
- Around line 61-69: The bot cache is using os.getenv("MODEL_NAME") only, which
diverges from configure_host()'s fallback to PRAISONAI_MODEL; update the Agent
creation in the _bots_cache block (where Agent(name="PraisonBot", ... llm=... )
is instantiated) to use the same lookup as configure_host(), e.g., check
MODEL_NAME first and fallback to PRAISONAI_MODEL (or reuse the same helper used
by configure_host()) so deployments that set only PRAISONAI_MODEL pick up the
correct model.
In `@src/praisonai/praisonai/ui_chat/default_app.py`:
- Around line 75-82: The Agent instantiation and the other legacy path use
inconsistent env-var lookups for the model name; unify model resolution by
centralizing the logic (e.g., add/reuse a single resolver like get_model_name()
or use configure_host()’s resolution) and replace direct os.getenv("MODEL_NAME",
...) and any os.getenv("PRAISONAI_MODEL", ...) checks with that resolver when
constructing Agent (llm parameter) and in the other referenced block (lines
around 119-121) so both legacy paths always pick the same model.
- Around line 56-60: on_settings currently only pops _agents_cache using the raw
session_id key but cache keys are formed as "{session_id}:{settings_key}", so
stale entries remain; update the decorated async function on_settings to iterate
over _agents_cache keys and remove all entries whose key starts with
f"{session_id}:" (use a snapshot of keys like list(_agents_cache.keys()) to
avoid mutating during iteration) so that all per-session cached agents are
cleared when settings change.
In `@src/praisonai/praisonai/ui_dashboard/default_app.py`:
- Around line 25-29: Update the agent_kwargs model lookup in default_app.py so
it follows the same precedence as other default apps: check
os.getenv("MODEL_NAME") first, then os.getenv("PRAISONAI_MODEL"), and finally
fall back to "gpt-4o-mini"; modify the "llm" value inside the agent_kwargs dict
accordingly (refer to agent_kwargs and its "llm" key) so processes configured
via MODEL_NAME receive the same model everywhere.
In `@src/praisonai/praisonai/ui_realtime/default_app.py`:
- Around line 71-79: The cached legacy realtime Agent is only reading MODEL_NAME
and ignoring the PRAISONAI_MODEL fallback used by configure_host(); update the
Agent creation in the _realtime_cache branch so its llm value uses the same
fallback chain (e.g., read MODEL_NAME, then PRAISONAI_MODEL, then the hardcoded
default) or call the same helper used by configure_host() to resolve the model
name; change the llm argument in Agent(...) inside the if session_id not in
_realtime_cache block accordingly so legacy realtime honors PRAISONAI_MODEL.
In `@src/praisonai/tests/integration/test_aiui_backends_injection.py`:
- Around line 10-19: The test test_build_host_injects_backends leaves host
bootstrap globals from previous tests causing order-dependent failures; fix by
resetting host bootstrap state before calling build_host_app: import
praisonai.integration.host_app as host_app and call a dedicated reset function
(e.g. host_app.reset_host_bootstrap() or host_app._reset_bootstrap()) in the
test right after backends.clear_backends(), and if that reset function does not
exist add one in host_app that reinitializes the module-level bootstrap globals
used by build_host_app.
In `@src/praisonai/tests/integration/test_aiui_gateway_parity.py`:
- Line 22: The assertion currently uses "or host_app._CONFIGURED" which lets the
test pass without verifying the hooks backend is present; change the test to
explicitly assert the hooks backend is wired by requiring "hooks" to be in
backends.list_backends() (and if a runtime fallback is needed, replace
host_app._CONFIGURED with an explicit check such as a
host_app.has_backend('hooks') or inspecting host_app._configured_backends) so
the test fails when the hooks backend is missing; update the assertion around
backends.list_backends() and host_app to perform an explicit presence check
rather than relying on host_app._CONFIGURED.
In `@src/praisonai/tests/integration/test_aiui_host_sse.py`:
- Around line 42-45: The SSE assertion is too lax—replace the loose check with
strict SSE validation: assert response.status_code == 200 and verify the
response headers contain 'text/event-stream', then ensure the streamed chunks
from response.iter_text() are non-empty and that at least one chunk (or the
joined body) contains lines beginning with 'data:' (e.g., each chunk or body
splitlines() has at least one line that startswith 'data:'). Update the
assertions around response.status_code, response.headers, response.iter_text(),
chunks and body in test_aiui_host_sse.py accordingly.
In `@src/praisonai/tests/integration/test_aiui_host.py`:
- Around line 80-86: The test_host_import_performance currently measures a warm
import because praisonai.integration.host_app is already cached; before starting
the timer, remove it from sys.modules (e.g. use
sys.modules.pop("praisonai.integration.host_app", None)) to force a cold import,
then start the timer and import praisonai.integration.host_app and assert the
elapsed time; update the test_host_import_performance function to perform this
pop so the measured import reflects cold-import performance.
---
Outside diff comments:
In `@src/praisonai-agents/praisonaiagents/session/README.md`:
- Line 165: Update the inconsistent session directory path in the README: change
the comment on the session_dir parameter (session_dir: Optional[str] = None, #
Default: ~/.praison/sessions/) to match the documented storage location used
elsewhere (e.g., ~/.praisonai/sessions/{session_id}.json). Ensure the text
referencing the session_dir parameter and any examples use the exact same path
string as the earlier documentation (the line that mentions
~/.praisonai/sessions/{session_id}.json) so both places consistently refer to
~/.praisonai/sessions/{session_id}.json.
In `@src/praisonai-agents/praisonaiagents/telemetry/protocols.py`:
- Around line 124-195: protocols.py currently contains concrete adapter
implementations (InMemoryUsageQuery, TokenCollectorUsageQuery) instead of only
interface contracts; move the adapter classes out of protocols.py into new
adapter modules (e.g., telemetry/adapters/in_memory_usage_query.py and
telemetry/adapters/token_collector_usage_query.py) preserving their methods
get_summary, list_recent, get_by_task, and get_by_agent, and leave protocols.py
containing only the Protocol/ABC definitions for the usage query interface;
update imports/exports so callers import the implementations from the new
adapter modules and the interface from protocols.py.
In `@src/praisonai/praisonai/__init__.py`:
- Around line 133-142: The module removed lazy resolution for AgentOS causing
"from praisonai import AgentOS" to fail; restore handling in __getattr__ by
adding a branch that returns the AgentOS symbol (importing AgentOS from .app)
when name == "AgentOS" so the exported name in __all__ is resolved lazily;
update the __getattr__ function to include this case alongside existing branches
like 'build_host_app' and 'configure_host' to preserve backward compatibility.
---
Minor comments:
In `@src/praisonai-agents/praisonaiagents/session/store.py`:
- Around line 527-529: The session summary construction incorrectly uses
Python's `or` which treats 0 as falsy and will drop valid zero values for fields
like `total_tokens` and `cost`; update the logic that sets `"model"`,
`"total_tokens"`, and `"cost"` (the dict entries currently using
`data.get("...") or ...`) to explicitly check for None (e.g., use conditional
expressions or `if x is not None else` chains) so that 0 is preserved but
None/falsy-missing values fall back to the other sources like
`data.get("token_count")` or `(data.get("metadata") or {}).get("total_tokens")`.
In `@src/praisonai/praisonai/integration/bridges/approvals_bridge.py`:
- Around line 26-29: The current return is leaking mutable internal state by
returning registry._requirements directly; change the "requirements" value to a
defensive copy (e.g., use copy.deepcopy(getattr(registry, "_requirements", {}))
or at minimum dict(getattr(registry, "_requirements", {}))) so callers cannot
mutate internals, and add the necessary import for copy if using deepcopy; keep
the rest of the returned keys (e.g., "auto_approve_env" /
registry.is_env_auto_approve()) unchanged.
In `@src/praisonai/praisonai/integration/host_app.py`:
- Around line 51-52: The current truthiness check "if modules:" prevents
honoring an explicit empty list; change the conditional around the
dashboard_opts assignment so it checks for None explicitly (e.g., use "if
modules is not None:"), then set dashboard_opts["modules"] = list(modules) as
before so callers can pass [] to clear modules while still ignoring a
missing/None value.
In `@src/praisonai/tests/integration/test_aiui_host_agentic.py`:
- Around line 22-31: The test uses a fixed session_id ("agentic-ac3-test")
causing collisions; change the test to generate a unique session_id per run
(e.g., using uuid.uuid4() or timestamp) before calling client.post("/run") so
the created session is guaranteed unique; update references to session_id that
follow (the client.post("/run") call, client.get("/sessions") check and the
ids/assertion logic) to use the generated value so the assertion reliably
verifies the newly created session.
---
Nitpick comments:
In `@src/praisonai-agents/praisonaiagents/agent/memory_mixin.py`:
- Around line 402-407: The code calls _persist_session_stats() for both user and
assistant turns causing unnecessary writes; change the logic in the try block
inside memory_mixin.py so that _persist_session_stats() is invoked only after
adding an assistant message (i.e., move the _persist_session_stats() call into
the elif role == "assistant" branch immediately after
self._session_store.add_assistant_message(self._session_id, content)), leaving
user handling (self._session_store.add_user_message) unchanged and preserving
the surrounding exception handling.
In `@src/praisonai-agents/praisonaiagents/agent/tool_execution.py`:
- Around line 293-304: Add an inline comment above the StreamEvent emission in
tool_execution.py where _Agent__stream_emitter.emit(StreamEvent(...,
type=StreamEventType.TOOL_CALL_END, ...)) is called to document the semantic
difference between TOOL_CALL_RESULT and TOOL_CALL_END: state that
TOOL_CALL_RESULT carries the tool's result payload while TOOL_CALL_END signals
lifecycle completion (no further events for this tool_call_id/function_name) and
is intended for UI finalization/cleanup; reference the
StreamEvent/StreamEventType types, the emit call, and the tool_call_id to make
the contract clear to future maintainers.
In `@src/praisonai-agents/praisonaiagents/scheduler/README.md`:
- Around line 1-30: Replace the plain Markdown in this README with Mintlify
components: convert the "Use when" bullet lists into a <Steps> component for
each of ScheduleRunner and ScheduleLoop, wrap the Python example in a
<CodeGroup> block, and add a <Note> (or Warning) callout that highlights the
stateless vs daemon distinction for ScheduleRunner and ScheduleLoop (mentioning
ScheduleRunner, ScheduleLoop, FileScheduleStore, and ensure_schedule_runner to
keep context). Keep the same headings and content but restructure into Mintlify
components for improved presentation and consistency with the project's docs
guidelines.
In `@src/praisonai-agents/praisonaiagents/session/README.md`:
- Around line 1-207: Convert the plain Markdown README sections into Mintlify
components: wrap the multiple Python examples in the "Quick Start" and "Advanced
Usage" sections using <CodeGroup>, turn the "Advanced Usage" block into an
<Accordion> with separate panels (Direct Session Store Access, Custom Session
Directory, Using with DB Adapter), mark Multi-Process Safety and Context Caching
as <Note> callouts, and render the "Session metadata contract" and "Behavior
Matrix" tables inside <Card> components; keep all API reference class names
(DefaultSessionStore, SessionData, SessionMessage) and headings intact while
replacing the corresponding Markdown code blocks and tables with the stated
Mintlify components for consistent formatting.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: a9be8b84-0cf2-4305-80e0-13179298b80d
📒 Files selected for processing (44)
src/praisonai-agents/praisonaiagents/agent/memory_mixin.pysrc/praisonai-agents/praisonaiagents/agent/tool_execution.pysrc/praisonai-agents/praisonaiagents/agents/agents.pysrc/praisonai-agents/praisonaiagents/hooks/query_protocol.pysrc/praisonai-agents/praisonaiagents/hooks/registry.pysrc/praisonai-agents/praisonaiagents/llm/llm.pysrc/praisonai-agents/praisonaiagents/scheduler/README.mdsrc/praisonai-agents/praisonaiagents/session/README.mdsrc/praisonai-agents/praisonaiagents/session/protocols.pysrc/praisonai-agents/praisonaiagents/session/store.pysrc/praisonai-agents/praisonaiagents/skills/protocols.pysrc/praisonai-agents/praisonaiagents/telemetry/protocols.pysrc/praisonai-agents/praisonaiagents/telemetry/token_collector.pysrc/praisonai-agents/praisonaiagents/telemetry/token_telemetry.pysrc/praisonai/docs/channels-gateway.mdsrc/praisonai/docs/pattern-d-platform.mdsrc/praisonai/praisonai/__init__.pysrc/praisonai/praisonai/claw/default_app.pysrc/praisonai/praisonai/cli/commands/dashboard.pysrc/praisonai/praisonai/cli/commands/ui.pysrc/praisonai/praisonai/integration/__init__.pysrc/praisonai/praisonai/integration/_legacy_handlers.pysrc/praisonai/praisonai/integration/bridges/__init__.pysrc/praisonai/praisonai/integration/bridges/approvals_bridge.pysrc/praisonai/praisonai/integration/bridges/hooks_query.pysrc/praisonai/praisonai/integration/bridges/schedules_runner.pysrc/praisonai/praisonai/integration/bridges/usage_bridge.pysrc/praisonai/praisonai/integration/bridges/workflows_service.pysrc/praisonai/praisonai/integration/context_files.pysrc/praisonai/praisonai/integration/gateway_host.pysrc/praisonai/praisonai/integration/host_app.pysrc/praisonai/praisonai/integration/pages/bot_health.pysrc/praisonai/praisonai/integration/pages/workflow_runs.pysrc/praisonai/praisonai/ui_agents/default_app.pysrc/praisonai/praisonai/ui_bot/default_app.pysrc/praisonai/praisonai/ui_chat/default_app.pysrc/praisonai/praisonai/ui_dashboard/default_app.pysrc/praisonai/praisonai/ui_realtime/default_app.pysrc/praisonai/tests/integration/test_aiui_backends_injection.pysrc/praisonai/tests/integration/test_aiui_gateway_parity.pysrc/praisonai/tests/integration/test_aiui_host.pysrc/praisonai/tests/integration/test_aiui_host_agentic.pysrc/praisonai/tests/integration/test_aiui_host_isolation.pysrc/praisonai/tests/integration/test_aiui_host_sse.py
| @runtime_checkable | ||
| class HooksQueryProtocol(Protocol): | ||
| """Read path for registered hooks.""" | ||
|
|
||
| def list_hooks_for_api(self) -> List[Dict]: | ||
| ... |
There was a problem hiding this comment.
🛠️ Refactor suggestion | 🟠 Major | ⚡ Quick win
Move this interface into hooks/protocols.py.
HooksQueryProtocol is the module's contract, so defining it in query_protocol.py breaks the required WHAT/HOW split for core SDK protocols.
As per coding guidelines, "All major modules in the core SDK must have a protocols.py file defining WHAT (interface contracts), with separate adapter files implementing HOW (concrete implementations)."
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/praisonai-agents/praisonaiagents/hooks/query_protocol.py` around lines 8
- 13, The HooksQueryProtocol interface (class HooksQueryProtocol with method
list_hooks_for_api) should be moved out of the query_protocol module into the
module that defines WHAT contracts (hooks.protocols): create the Protocol class
there, preserve `@runtime_checkable` and the method signature, then update any
imports in code that currently import HooksQueryProtocol from query_protocol to
import it from hooks.protocols and remove the old definition from
query_protocol; ensure typing imports (Protocol, List, Dict) remain present and
adjust package exports if necessary.
| def list_skills_for_api() -> List[Dict[str, Any]]: | ||
| """Default catalog adapter using SkillManager when available.""" | ||
| try: | ||
| from .manager import SkillManager | ||
|
|
||
| manager = SkillManager() | ||
| return [ | ||
| { | ||
| "name": s.name, | ||
| "description": s.description, | ||
| "location": getattr(s, "location", ""), | ||
| } | ||
| for s in manager.get_available_skills() | ||
| ] | ||
| except Exception: | ||
| return [] |
There was a problem hiding this comment.
🛠️ Refactor suggestion | 🟠 Major | ⚡ Quick win
Keep the default catalog adapter out of protocols.py.
list_skills_for_api() is a concrete SkillManager-backed adapter, so it should live in a separate adapter/helper module with protocols.py reserved for interface contracts.
As per coding guidelines, "All major modules in the core SDK must have a protocols.py file defining WHAT (interface contracts), with separate adapter files implementing HOW (concrete implementations)."
🧰 Tools
🪛 Ruff (0.15.13)
[warning] 192-192: Do not catch blind exception: Exception
(BLE001)
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/praisonai-agents/praisonaiagents/skills/protocols.py` around lines 178 -
193, The function list_skills_for_api is a concrete SkillManager-backed adapter
and should be moved out of protocols.py; create a new adapter module (e.g.,
skills_adapter or skill_catalog_adapter) that implements list_skills_for_api by
importing SkillManager and calling manager.get_available_skills(), keep
protocols.py limited to interface definitions only (remove list_skills_for_api
from protocols.py), update any imports to reference the new adapter module, and
ensure the adapter returns the same dict shape ("name", "description",
"location") as before.
| try: | ||
| from .manager import SkillManager | ||
|
|
||
| manager = SkillManager() | ||
| return [ | ||
| { | ||
| "name": s.name, | ||
| "description": s.description, | ||
| "location": getattr(s, "location", ""), | ||
| } | ||
| for s in manager.get_available_skills() | ||
| ] | ||
| except Exception: | ||
| return [] |
There was a problem hiding this comment.
Don't collapse every catalog failure into an empty list.
Catching Exception here will hide real discovery/runtime bugs and make the API report "no skills" when the catalog is actually broken. If graceful degradation is intentional, limit it to import failures and let actual catalog errors surface.
Suggested fix
def list_skills_for_api() -> List[Dict[str, Any]]:
"""Default catalog adapter using SkillManager when available."""
try:
from .manager import SkillManager
@@
}
for s in manager.get_available_skills()
]
- except Exception:
+ except ImportError:
return []📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| try: | |
| from .manager import SkillManager | |
| manager = SkillManager() | |
| return [ | |
| { | |
| "name": s.name, | |
| "description": s.description, | |
| "location": getattr(s, "location", ""), | |
| } | |
| for s in manager.get_available_skills() | |
| ] | |
| except Exception: | |
| return [] | |
| try: | |
| from .manager import SkillManager | |
| manager = SkillManager() | |
| return [ | |
| { | |
| "name": s.name, | |
| "description": s.description, | |
| "location": getattr(s, "location", ""), | |
| } | |
| for s in manager.get_available_skills() | |
| ] | |
| except ImportError: | |
| return [] |
🧰 Tools
🪛 Ruff (0.15.13)
[warning] 192-192: Do not catch blind exception: Exception
(BLE001)
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/praisonai-agents/praisonaiagents/skills/protocols.py` around lines 180 -
193, The current broad except Exception hides real discovery/runtime errors;
restrict the graceful fallback to import failures only by isolating the "from
.manager import SkillManager" in its own try/except ImportError block (or
ModuleNotFoundError) and return [] only on that import error, then instantiate
SkillManager and call manager.get_available_skills() without swallowing
exceptions so runtime/catalog errors surface; reference the SkillManager import,
the manager variable, and manager.get_available_skills() when making this
change.
| async def legacy_reply(message: str, session_id: str = "default"): | ||
| from praisonaiagents import Agent | ||
|
|
||
| agent = Agent( | ||
| name=agent_name, | ||
| instructions="You are a helpful assistant.", | ||
| llm="gpt-4o-mini", | ||
| ) | ||
| return agent.run(message) |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# Verify Agent API shape (run/arun sync-vs-async) in local sources
rg -nP --type=py -C2 'class\s+Agent\b|^\s*(async\s+def|def)\s+(run|arun)\s*\(' src/praisonai-agents/praisonaiagents
# Verify the handler call site
rg -nP --type=py -C2 '`@aiui`\.reply|async\s+def\s+legacy_reply|agent\.run\(' src/praisonai/praisonai/integration/_legacy_handlers.pyRepository: MervinPraison/PraisonAI
Length of output: 18750
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "=== _legacy_handlers.py (context) ==="
sed -n '1,120p' src/praisonai/praisonai/integration/_legacy_handlers.py
echo
echo "=== execution_mixin.py (arun) ==="
sed -n '90,170p' src/praisonai-agents/praisonaiagents/agent/execution_mixin.py
echo
echo "=== execution_mixin.py (run) ==="
sed -n '190,280p' src/praisonai-agents/praisonaiagents/agent/execution_mixin.pyRepository: MervinPraison/PraisonAI
Length of output: 7785
Avoid blocking the event loop in legacy_reply; use await agent.arun(message) instead of agent.run()
Agent.run() is synchronous in praisonaiagents (it calls the non-async chat path), so calling it inside async def legacy_reply can block the event loop. Agent.arun() exists and is the correct async alternative.
Proposed fix
+import asyncio
@@
`@aiui.reply`
async def legacy_reply(message: str, session_id: str = "default"):
from praisonaiagents import Agent
@@
- return agent.run(message)
+ if hasattr(agent, "arun"):
+ return await agent.arun(message)
+ return await asyncio.to_thread(agent.run, message)📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| async def legacy_reply(message: str, session_id: str = "default"): | |
| from praisonaiagents import Agent | |
| agent = Agent( | |
| name=agent_name, | |
| instructions="You are a helpful assistant.", | |
| llm="gpt-4o-mini", | |
| ) | |
| return agent.run(message) | |
| async def legacy_reply(message: str, session_id: str = "default"): | |
| from praisonaiagents import Agent | |
| agent = Agent( | |
| name=agent_name, | |
| instructions="You are a helpful assistant.", | |
| llm="gpt-4o-mini", | |
| ) | |
| if hasattr(agent, "arun"): | |
| return await agent.arun(message) | |
| return await asyncio.to_thread(agent.run, message) |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/praisonai/praisonai/integration/_legacy_handlers.py` around lines 10 -
18, The async function legacy_reply currently calls the synchronous Agent.run
which can block the event loop; update legacy_reply to call and await the
asynchronous method Agent.arun(message) instead (i.e., replace
agent.run(message) with await agent.arun(message)) so the handler remains
non-blocking; ensure the created agent object (Agent(...)) supports arun and
propagate await back to the caller.
| if session_id not in _realtime_cache: | ||
| _realtime_cache[session_id] = Agent( | ||
| name="RealtimeAssistant", | ||
| instructions=( | ||
| "You are a voice-optimized assistant. " | ||
| "Keep responses conversational and concise for voice interaction." | ||
| ), | ||
| llm=os.getenv("MODEL_NAME", "gpt-4o-mini"), | ||
| ) |
There was a problem hiding this comment.
Use the same env fallback in the legacy realtime agent.
configure_host() accepts MODEL_NAME falling back to PRAISONAI_MODEL, but the cached realtime agent only checks MODEL_NAME. If only PRAISONAI_MODEL is set, legacy realtime will ignore it.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/praisonai/praisonai/ui_realtime/default_app.py` around lines 71 - 79, The
cached legacy realtime Agent is only reading MODEL_NAME and ignoring the
PRAISONAI_MODEL fallback used by configure_host(); update the Agent creation in
the _realtime_cache branch so its llm value uses the same fallback chain (e.g.,
read MODEL_NAME, then PRAISONAI_MODEL, then the hardcoded default) or call the
same helper used by configure_host() to resolve the model name; change the llm
argument in Agent(...) inside the if session_id not in _realtime_cache block
accordingly so legacy realtime honors PRAISONAI_MODEL.
| def test_build_host_injects_backends(): | ||
| import praisonaiui.backends as backends | ||
| from praisonai.integration.host_app import build_host_app | ||
|
|
||
| backends.clear_backends() | ||
| app = build_host_app(pages=["chat"]) | ||
| assert app is not None | ||
| registered = backends.list_backends() | ||
| assert "hooks" in registered | ||
| assert "workflows" in registered |
There was a problem hiding this comment.
Reset host bootstrap state to avoid order-dependent failures.
This test clears backend registry but does not reset host bootstrap globals, so previous tests can make this pass/fail nondeterministically.
Suggested fix
-def test_build_host_injects_backends():
+def test_build_host_injects_backends(monkeypatch):
import praisonaiui.backends as backends
+ import praisonaiui.server as srv
from praisonai.integration.host_app import build_host_app
+ from praisonai.integration import host_app
+ host_app._CONFIGURED = False
+ srv._provider = None
backends.clear_backends()
app = build_host_app(pages=["chat"])
assert app is not None📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| def test_build_host_injects_backends(): | |
| import praisonaiui.backends as backends | |
| from praisonai.integration.host_app import build_host_app | |
| backends.clear_backends() | |
| app = build_host_app(pages=["chat"]) | |
| assert app is not None | |
| registered = backends.list_backends() | |
| assert "hooks" in registered | |
| assert "workflows" in registered | |
| def test_build_host_injects_backends(monkeypatch): | |
| import praisonaiui.backends as backends | |
| import praisonaiui.server as srv | |
| from praisonai.integration.host_app import build_host_app | |
| from praisonai.integration import host_app | |
| host_app._CONFIGURED = False | |
| srv._provider = None | |
| backends.clear_backends() | |
| app = build_host_app(pages=["chat"]) | |
| assert app is not None | |
| registered = backends.list_backends() | |
| assert "hooks" in registered | |
| assert "workflows" in registered |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/praisonai/tests/integration/test_aiui_backends_injection.py` around lines
10 - 19, The test test_build_host_injects_backends leaves host bootstrap globals
from previous tests causing order-dependent failures; fix by resetting host
bootstrap state before calling build_host_app: import
praisonai.integration.host_app as host_app and call a dedicated reset function
(e.g. host_app.reset_host_bootstrap() or host_app._reset_bootstrap()) in the
test right after backends.clear_backends(), and if that reset function does not
exist add one in host_app that reinitializes the module-level bootstrap globals
used by build_host_app.
|
|
||
| host_app.configure_host(pages=["chat"]) | ||
| assert srv.get_datastore() is not None | ||
| assert "hooks" in backends.list_backends() or host_app._CONFIGURED |
There was a problem hiding this comment.
Tighten the parity assertion to verify actual backend wiring.
or host_app._CONFIGURED makes this pass even when hooks backend is missing.
Suggested fix
- assert "hooks" in backends.list_backends() or host_app._CONFIGURED
+ assert host_app._CONFIGURED is True
+ assert "hooks" in backends.list_backends()📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| assert "hooks" in backends.list_backends() or host_app._CONFIGURED | |
| assert host_app._CONFIGURED is True | |
| assert "hooks" in backends.list_backends() |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/praisonai/tests/integration/test_aiui_gateway_parity.py` at line 22, The
assertion currently uses "or host_app._CONFIGURED" which lets the test pass
without verifying the hooks backend is present; change the test to explicitly
assert the hooks backend is wired by requiring "hooks" to be in
backends.list_backends() (and if a runtime fallback is needed, replace
host_app._CONFIGURED with an explicit check such as a
host_app.has_backend('hooks') or inspecting host_app._configured_backends) so
the test fails when the hooks backend is missing; update the assertion around
backends.list_backends() and host_app to perform an explicit presence check
rather than relying on host_app._CONFIGURED.
| assert response.status_code == 200 | ||
| chunks = list(response.iter_text()) | ||
| body = "".join(chunks) | ||
| assert "data:" in body or len(chunks) > 0 |
There was a problem hiding this comment.
SSE validation is too permissive and can pass on non-SSE output.
"data:" in body or len(chunks) > 0 accepts arbitrary chunked responses.
Suggested fix
) as response:
assert response.status_code == 200
+ assert "text/event-stream" in response.headers.get("content-type", "")
chunks = list(response.iter_text())
body = "".join(chunks)
- assert "data:" in body or len(chunks) > 0
+ assert len(chunks) > 0
+ assert "data:" in body📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| assert response.status_code == 200 | |
| chunks = list(response.iter_text()) | |
| body = "".join(chunks) | |
| assert "data:" in body or len(chunks) > 0 | |
| assert response.status_code == 200 | |
| assert "text/event-stream" in response.headers.get("content-type", "") | |
| chunks = list(response.iter_text()) | |
| body = "".join(chunks) | |
| assert len(chunks) > 0 | |
| assert "data:" in body |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/praisonai/tests/integration/test_aiui_host_sse.py` around lines 42 - 45,
The SSE assertion is too lax—replace the loose check with strict SSE validation:
assert response.status_code == 200 and verify the response headers contain
'text/event-stream', then ensure the streamed chunks from response.iter_text()
are non-empty and that at least one chunk (or the joined body) contains lines
beginning with 'data:' (e.g., each chunk or body splitlines() has at least one
line that startswith 'data:'). Update the assertions around
response.status_code, response.headers, response.iter_text(), chunks and body in
test_aiui_host_sse.py accordingly.
| def test_host_import_performance(self): | ||
| start = time.perf_counter() | ||
| import praisonai.integration.host_app # noqa: F401 | ||
|
|
||
| elapsed_ms = (time.perf_counter() - start) * 1000 | ||
| assert elapsed_ms < 2000, f"host_app import took {elapsed_ms:.0f}ms" | ||
|
|
There was a problem hiding this comment.
🧩 Analysis chain
🌐 Web query:
In CPython, after a module is first imported, does a later import moduleresolve fromsys.modulescache, and should cold import timing tests remove the module fromsys.modules first?
💡 Result:
Yes. In CPython/Python, a later import module consults the module cache in sys.modules first; if the module name is present there with a non-None value, the import completes by using the existing module object from the cache (it does not re-run the module’s top-level code). This means “warm” import timing tests will mostly measure the cache lookup, not a full reload from disk [1]. For cold-import timing tests, you generally should ensure the module is not present in sys.modules before timing, otherwise you are not measuring a cold import. Python explicitly documents that sys.modules is writable and that deleting an entry invalidates the cache entry so the next import searches anew [1]. However, for accuracy you should note two important details: 1) Deleting the sys.modules key may not actually “destroy” the module object if other references to it still exist, so global state in the module can remain alive via those references [1]. 2) If an import fails, CPython removes only the failing module from sys.modules; modules that were already in the cache or imported as side-effects remain cached [1]. So your timing harness should start from a clean state (and ideally avoid import failures) to be truly “cold.” So the recommended approach for cold timing is: remove the target name from sys.modules (and, if you care about package parents/submodules, ensure relevant related entries like package/package.submodule aren’t cached either), then time the import. Citations: sys.modules is the first place checked and controls whether the import completes using an existing module object [1]. Deleting/invalidation of sys.modules entries causes the next import to search anew [1]. If loading fails, only the failing module is removed while other cached/side-effect modules remain [1].
Citations:
Fix the import timing test to measure cold-import performance.
CPython reuses already-imported modules from sys.modules, so because praisonai.integration.host_app is imported earlier in this file, this assertion largely measures a warm-cache lookup rather than true import cost. Remove the module from sys.modules (e.g., sys.modules.pop("praisonai.integration.host_app", None)) before starting the timer and then time the import again.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/praisonai/tests/integration/test_aiui_host.py` around lines 80 - 86, The
test_host_import_performance currently measures a warm import because
praisonai.integration.host_app is already cached; before starting the timer,
remove it from sys.modules (e.g. use
sys.modules.pop("praisonai.integration.host_app", None)) to force a cold import,
then start the timer and import praisonai.integration.host_app and assert the
elapsed time; update the test_host_import_performance function to perform this
pop so the measured import reflects cold-import performance.
Agent-Logs-Url: https://github.com/MervinPraison/PraisonAI/sessions/c2c4c0e0-0cb7-4608-ab5e-3a259f9bd635 Co-authored-by: MervinPraison <454862+MervinPraison@users.noreply.github.com>
Implemented and pushed targeted fixes in
Also added focused tests:
Validation:
CI check: the failed Copilot review workflow run was due to external weekly rate-limit (HTTP 429), not code/test failure. |
| def list_recent(self, limit: int = 50) -> List[Dict[str, Any]]: | ||
| return self._sink.records[-limit:] |
There was a problem hiding this comment.
-0 == 0, so records[-0:] evaluates to records[0:] and returns the entire list instead of an empty list. Any caller that passes limit=0 — for example a pagination boundary check — will receive the full usage history rather than zero records.
| def list_recent(self, limit: int = 50) -> List[Dict[str, Any]]: | |
| return self._sink.records[-limit:] | |
| def list_recent(self, limit: int = 50) -> List[Dict[str, Any]]: | |
| if limit == 0: | |
| return [] | |
| return self._sink.records[-limit:] |
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
src/praisonai-agents/praisonaiagents/agent/tool_execution.py (1)
309-320:⚠️ Potential issue | 🟠 Major | ⚡ Quick winEmit
TOOL_CALL_ENDon failure paths too.
TOOL_CALL_ENDis only sent on success. When tool execution raises, stream consumers won’t receive a terminal event and may keep pending UI state indefinitely. Emit end-event in theexceptpath (or move to afinally) with error metadata.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai-agents/praisonaiagents/agent/tool_execution.py` around lines 309 - 320, The code currently only calls _stream_emitter.emit(StreamEvent(..., type=StreamEventType.TOOL_CALL_END, ...)) on the successful path, so ensure TOOL_CALL_END is emitted on failure as well by moving the emit into a finally block (or duplicating it into the except) inside the function that runs tools (refer to _stream_emitter.emit, StreamEvent, StreamEventType.TOOL_CALL_END, tool_call_id, function_name, arguments, result_summary); include error metadata (e.g., an "error" field or "error_message" and a flag like "failed": true) when emitting from the except path so consumers can differentiate terminal success vs failure and always receive the terminal event.
🧹 Nitpick comments (1)
src/praisonai-agents/tests/unit/streaming/test_tool_call_events.py (1)
230-243: ⚡ Quick winAdd a behavior-level test for
TOOL_CALL_ENDin execution flow.This test validates emitter lookup, but it doesn’t verify the new end-event contract from tool execution. Please add one assertion path that executes a tool and checks
TOOL_CALL_START -> TOOL_CALL_RESULT -> TOOL_CALL_ENDordering and ID propagation.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/praisonai-agents/tests/unit/streaming/test_tool_call_events.py` around lines 230 - 243, Extend the test in TestToolExecutionMixinStreamEmitterLookup to actually execute a tool via the ToolExecutionMixin and assert the emitter emits TOOL_CALL_START then TOOL_CALL_RESULT then TOOL_CALL_END in that order and that the same call id propagates across events; specifically, obtain the StreamEventEmitter via agent._get_existing_stream_emitter(), perform the tool execution path on the CustomAgent (use the mixin method that triggers tool execution on your agent, e.g., the method that normally produces start/result/end events), capture emitted events from the emitter, assert the sequence of event.type equals praisonaiagents.streaming.events.TOOL_CALL_START, TOOL_CALL_RESULT, TOOL_CALL_END, and assert each event carries the same call id (e.g., event.data["id"] or event.id) across the three events, adding minimal setup/mocks needed to run the execution.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Outside diff comments:
In `@src/praisonai-agents/praisonaiagents/agent/tool_execution.py`:
- Around line 309-320: The code currently only calls
_stream_emitter.emit(StreamEvent(..., type=StreamEventType.TOOL_CALL_END, ...))
on the successful path, so ensure TOOL_CALL_END is emitted on failure as well by
moving the emit into a finally block (or duplicating it into the except) inside
the function that runs tools (refer to _stream_emitter.emit, StreamEvent,
StreamEventType.TOOL_CALL_END, tool_call_id, function_name, arguments,
result_summary); include error metadata (e.g., an "error" field or
"error_message" and a flag like "failed": true) when emitting from the except
path so consumers can differentiate terminal success vs failure and always
receive the terminal event.
---
Nitpick comments:
In `@src/praisonai-agents/tests/unit/streaming/test_tool_call_events.py`:
- Around line 230-243: Extend the test in
TestToolExecutionMixinStreamEmitterLookup to actually execute a tool via the
ToolExecutionMixin and assert the emitter emits TOOL_CALL_START then
TOOL_CALL_RESULT then TOOL_CALL_END in that order and that the same call id
propagates across events; specifically, obtain the StreamEventEmitter via
agent._get_existing_stream_emitter(), perform the tool execution path on the
CustomAgent (use the mixin method that triggers tool execution on your agent,
e.g., the method that normally produces start/result/end events), capture
emitted events from the emitter, assert the sequence of event.type equals
praisonaiagents.streaming.events.TOOL_CALL_START, TOOL_CALL_RESULT,
TOOL_CALL_END, and assert each event carries the same call id (e.g.,
event.data["id"] or event.id) across the three events, adding minimal
setup/mocks needed to run the execution.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 7670462b-53e7-4299-9175-d160531e8622
📒 Files selected for processing (4)
src/praisonai-agents/praisonaiagents/agent/tool_execution.pysrc/praisonai-agents/praisonaiagents/telemetry/protocols.pysrc/praisonai-agents/tests/unit/streaming/test_tool_call_events.pysrc/praisonai-agents/tests/unit/telemetry/test_token_usage_sink.py
|
@claude validate every file step by step as there are too many file changes. Validate indpeth make sure all working and all tests passed. if any issues fix. It shoudl follow agents.md also make sure it shouldn't bloat the core sdk for the sake of it |
|
Claude finished @MervinPraison's task —— View job Final Comprehensive Validation Report ✅I have conducted a thorough file-by-file validation of all 46 changed files in this PR against AGENTS.md guidelines. ✅ Architecture Compliance: PASSEDCore SDK (praisonaiagents) Analysis:
Wrapper (praisonai) Analysis:
✅ Fixed Critical Architecture Violations:
✅ Fixed Code Quality Issues:
✅ Backward Compatibility: MAINTAINED
✅ Performance: NO REGRESSIONS
🎯 Final Verdict: APPROVE FOR MERGEThis PR successfully implements PraisonAI ↔ PraisonAIUI integration following all AGENTS.md architectural guidelines:
Recommendation: This PR is architecturally sound and ready for merge after the architecture fixes I've implemented. Commits made: |
Summary
praisonai.integrationhost bootstrap (configure_host,build_host_app,run_integrated_gateway) wiring PraisonAISessionDataStore, PraisonAIProvider, and optional L1 bridges into praisonaiui viaset_backend().default_app.pymodules plus newui_dashboardto use the shared host;praisonai dashboard --aiuinow runs an in-process Pattern B host.get_token_collector(), session metadata persistence,TOOL_CALL_ENDstreaming, and 16 integration tests.Test plan
pytest src/praisonai/tests/integration/test_aiui_host*.py src/praisonai/tests/integration/test_aiui_gateway_parity.py src/praisonai/tests/integration/test_aiui_backends_injection.py(16 passed)praisonai dashboard --aiuismoke on local machineOPENAI_API_KEY(test_aiui_host_agentic.py)feat/l1-integration-backends) for full end-to-end UINotes
PRAISONAI_HOST_LEGACY=1to skip provider wiring (callback-only mode).from praisonai.integration import run_integrated_gatewayor lazyfrom praisonai import build_host_app.Summary by CodeRabbit
New Features
Documentation
Tests