Probe Agent Overview

February 18, 2026 · View on GitHub

Probe Agent is an AI-powered code exploration and interaction platform built on Node.js. It combines Probe's semantic code search capabilities with intelligent agent orchestration, enabling you to build sophisticated AI coding assistants.

What is Probe Agent?

Probe Agent provides:

  • AI-Powered Code Understanding: Leverage LLMs to explore, explain, and modify code
  • Multi-Provider Support: Works with Anthropic Claude, OpenAI GPT, Google Gemini, AWS Bedrock, Claude Code, and Codex
  • Tool Orchestration: Automatic tool calling for search, extraction, editing, and more
  • Session Management: Persistent conversations with history and context compaction
  • Protocol Support: MCP and ACP for integration with AI editors and agent systems

TL;DR - Quick Example

import { ProbeAgent } from '@probelabs/probe/agent';

const agent = new ProbeAgent({
  path: './my-project',
  provider: 'anthropic'
});

await agent.initialize();
const response = await agent.answer("How does authentication work in this codebase?");
console.log(response);

Agent Flavors

Probe Agent comes in several forms to suit different use cases:

1. ProbeAgent SDK

The core Node.js class for programmatic integration:

npm install @probelabs/probe

import { ProbeAgent } from '@probelabs/probe/agent';

Use when: Building custom AI tools, integrating into existing applications, or creating specialized workflows.

2. Probe Chat CLI

Interactive terminal-based chat interface:

npx -y @probelabs/probe-chat@latest ./my-project

Use when: Exploring codebases interactively, quick code questions, or developer productivity.

3. Probe Chat Web

Browser-based chat interface with syntax highlighting:

npx -y @probelabs/probe-chat@latest --web ./my-project

Use when: Team collaboration, sharing code insights, or when you prefer a visual interface.

4. MCP Server

Model Context Protocol server for AI editor integration:

probe mcp

Use when: Integrating with Cursor, Windsurf, Claude Code, or other MCP-compatible AI editors.

5. ACP Server

Agent Communication Protocol for advanced agent orchestration:

node index.js --acp

Use when: Building multi-agent systems, complex workflows, or custom AI platforms.

Key Capabilities

Probe Agent uses semantic search to find relevant code:

const agent = new ProbeAgent({ path: './src' });
await agent.initialize();

// The agent automatically uses search tools
const response = await agent.answer("Find all functions that handle user authentication");

Code Extraction & Context

Extract full function and class definitions with context:

const response = await agent.answer("Extract the login function and explain how it works");

Search inside your project's dependencies - npm packages, Go modules, and Rust crates:

// Ask the agent to search inside dependencies
const response = await agent.answer(
  "How does createAnthropic work in the @ai-sdk/anthropic package?"
);

// The agent uses special path prefixes:
// - js:package-name (npm packages)
// - go:module/path (Go modules)
// - rust:crate-name (Rust crates)

Example prompts:

  • "Search for how express handles middleware"
  • "Look inside the gin library to see how routing works"
  • "Find the Serialize trait definition in serde"

Code Editing (Optional)

Enable code modifications with safeguards:

const agent = new ProbeAgent({
  path: './src',
  allowEdit: true,
  enableBash: true
});

Multi-Provider Resilience

Automatic retry and fallback across providers:

const agent = new ProbeAgent({
  provider: 'anthropic',
  retry: { maxRetries: 3, backoffFactor: 2 },
  fallback: { strategy: 'any' }
});

Architecture Overview

┌─────────────────────────────────────────────────────────────┐
│                      Your Application                        │
└─────────────────────────────┬───────────────────────────────┘
                              │
┌─────────────────────────────▼───────────────────────────────┐
│                       ProbeAgent SDK                         │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────────────┐  │
│  │   Engines   │  │    Tools    │  │  Session Manager    │  │
│  │  (Claude,   │  │  (Search,   │  │  (History, Tokens,  │  │
│  │   OpenAI,   │  │   Extract,  │  │   Compaction)       │  │
│  │   Gemini)   │  │   Edit...)  │  │                     │  │
│  └──────┬──────┘  └──────┬──────┘  └──────────┬──────────┘  │
└─────────┼────────────────┼────────────────────┼─────────────┘
          │                │                    │
┌─────────▼────────────────▼────────────────────▼─────────────┐
│                      Probe CLI (Rust)                        │
│         Semantic Search • AST Parsing • Code Extraction      │
└─────────────────────────────────────────────────────────────┘

Supported Providers

ProviderTypeConfiguration
Anthropic ClaudeAPIANTHROPIC_API_KEY
OpenAI GPTAPIOPENAI_API_KEY
Google GeminiAPIGOOGLE_GENERATIVE_AI_API_KEY
AWS BedrockAPIAWS credentials
Claude CodeCLIRequires claude command
CodexCLIRequires codex command

Feature Matrix

FeatureSDKChat CLIChat WebMCPACP
Code Search✓✓✓✓✓
Dependency Search✓✓✓✓✓
Code Extraction✓✓✓✓✓
AST Queries✓✓✓✓✓
Code Editing✓✓✓-✓
Bash Execution✓✓✓--
Session Persistence✓✓✓-✓
Multi-Provider✓✓✓-✓
Streaming✓✓✓-✓
Token Tracking✓✓✓-✓
Delegation✓---✓

When to Use What

Use ProbeAgent SDK when:

  • Building custom AI applications
  • Integrating into existing Node.js projects
  • Need full control over agent behavior
  • Creating specialized workflows

Use Probe Chat when:

  • Interactive code exploration
  • Quick questions about a codebase
  • Developer productivity
  • Team code reviews

Use MCP Server when:

  • Working with AI code editors (Cursor, Windsurf)
  • Want Probe search in Claude Code
  • Need standardized tool integration

Use ACP Server when:

  • Building multi-agent systems
  • Need session management and isolation
  • Creating custom AI platforms
  • Complex orchestration requirements

Next Steps