Technical terms and concepts used throughout Probe documentation.
An advanced protocol for agent-to-agent communication in Probe. Provides session management, streaming, and structured tool execution. See ACP Protocol.
A tree representation of source code structure. Probe uses tree-sitter to parse code into ASTs for precise structural queries. For example, fn $NAME() { $$$BODY } matches Rust function definitions.
A structural code search tool that uses AST patterns. Probe integrates ast-grep for the query command.
A ranking algorithm used for information retrieval. BM25 is Probe's default reranker, balancing term frequency and document length for relevance scoring.
A contiguous section of code extracted by Probe. Blocks typically correspond to functions, classes, or other semantic units. Adjacent blocks within --merge-threshold lines are merged.
Anthropic's AI coding assistant CLI. Probe integrates with Claude Code as both an MCP server and as a backend for the implement tool.
OpenAI's legacy code generation CLI. Probe supports Codex as an alternative engine for code generation tasks. Note: OpenAI's Codex API was deprecated in 2023; modern usage typically involves GPT-4 models instead.
The process of reducing conversation history to fit within token limits while preserving essential information. See Context Compaction.
The maximum number of tokens an AI model can process in a single request. Probe's --max-tokens flag helps stay within these limits.
A ProbeAgent feature that distributes tasks to specialized subagents. Useful for complex multi-step operations. See Delegation.
Query syntax supported by Probe's search command. Includes operators like AND, OR, NOT, wildcards (*), and exact phrases ("...").
The AI provider/model combination used by ProbeAgent. Supported engines include Anthropic Claude, OpenAI GPT, Google Gemini, AWS Bedrock, Claude Code, and Codex.
Probe command that retrieves complete code blocks from files by location (line number) or symbol name.
A resilience strategy where ProbeAgent automatically tries alternative providers when the primary fails. See Retry & Fallback.
A pattern matching syntax for file paths. Examples: *.ts (TypeScript files), **/*.test.js (all test files).
A callback function triggered by ProbeAgent lifecycle events. Hooks enable custom behavior for tool execution, message handling, and more. See Hooks System.
A ranking algorithm combining BM25 and TF-IDF for improved relevance. Use --reranker hybrid or --reranker hybrid2.
The messaging protocol used by ACP. A lightweight remote procedure call protocol encoded in JSON.
A standard protocol for AI tool integration. Probe implements MCP for integration with AI editors like Cursor and Claude Code. See MCP Protocol.
The --merge-threshold option controls how many lines apart code blocks can be while still being merged into a single result. Default is 5 lines.
Probe's default output format showing hierarchical code structure with symbols, line numbers, and brief descriptions.
Probe's thread pool for tree-sitter parsers. Configurable via PROBE_PARSER_POOL_SIZE environment variable.
The Node.js SDK component of Probe for building AI coding assistants. Includes ProbeAgent class, tools, MCP/ACP support.
The Rust-based command-line tool for semantic code search. Includes search, extract, query, and grep commands.
The main class in Probe's Node.js SDK for creating AI agents with code search capabilities.
Probe command for AST-based structural code search. Uses tree-sitter patterns like fn $NAME() { $$$BODY }.
A Rust library for parallel processing. Probe uses Rayon for concurrent file scanning and processing.
An algorithm that scores and orders search results. Options include bm25, tfidf, hybrid, and BERT-based rerankers.
A fast line-oriented search tool. Probe uses ripgrep (via the ignore crate) for initial file scanning.
Automatic retry of failed API requests with exponential backoff. See Retry & Fallback.
Query filters in Probe's search syntax. Examples: ext:ts (file extension), lang:python (language), dir:tests (directory).
A conversation context in ProbeAgent. Sessions maintain history, token tracking, and state across multiple interactions.
CPU-level parallelization for operations like tokenization and ranking. Probe uses SIMD for performance optimization.
Discoverable agent capabilities in ProbeAgent. Skills can be listed and executed dynamically.
An interface for persisting ProbeAgent conversation history. Default is InMemoryStorageAdapter.
Linguistic normalization that reduces words to their root form. "running" → "run". Enabled by default in Probe search.
A ranking algorithm that weighs term importance by frequency and rarity. Available via --reranker tfidf.
A unit of text processing for AI models. Roughly 4 characters in English. Probe tracks tokens for context window management.
A ProbeAgent class for tracking token usage across requests.
A capability that AI agents can invoke. Probe provides tools like search, query, extract, edit, bash.
A parser generator for programming languages. Probe uses tree-sitter for AST parsing and structural code analysis.
A framework for building AI applications. ProbeAgent uses Vercel AI SDK for model integration.
The common parent directory for all paths in a ProbeAgent session. Computed from allowedFolders.
The format ProbeAgent uses for tool calls in conversations. Tools are invoked via XML tags like <search><query>...</query></search>.
- FAQ - Frequently asked questions
- Troubleshooting - Common issues
- Environment Variables - Configuration reference