intutic

mcp
Guvenlik Denetimi
Uyari
Health Uyari
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Uyari
  • process.env — Environment variable access in apps/docs/.vitepress/config.ts
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

The open source circuit breaker for AI agents. Real-time security, secret DLP, and loop burn prevention for Claude Code, Cursor, Antigravity, LangGraph, n8n and many more.

README.md

Intutic — The Circuit Breaker for AI Agents

Real-time security, secret DLP, and loop burn prevention for autonomous AI coding agents.

GitHub Stars
License: MIT
Documentation
Build Status
PRs Welcome

QuickstartArchitectureKey FeaturesSupported HarnessesDocs


Intutic Hero Demo — Real-time AI Agent Circuit Breaker

💡 Why Intutic?

Existing AI observability tools (like LangSmith or Portkey) are passive. They record execution logs after an agent leaks a secret, deletes files, or loops into hundreds of dollars of API spend.

Intutic is an active, low-latency circuit breaker. It sits in the tool-call path between your AI agents and local shell/production APIs. Every tool execution passes through a sub-5ms policy evaluation chain — blocking dangerous commands before they run and steering agentic loops in real time.


🏗️ Architecture

Intutic runs as a high-performance local or self-hosted proxy (written in Rust) alongside a lightweight bidirectional config sync daemon (sync-daemon):

flowchart TD
    subgraph DevEnvironment[" 💻 Developer Environment "]
        Agent["🤖 AI Coding Agent<br><i>(Claude Code, Cursor, Aider, LangGraph)</i>"]
        SOP["📝 Local SOP Rules<br><i>(CLAUDE.md / .cursorrules / SKILL.md)</i>"]
    end

    subgraph HotPathProxy[" ⚡ Intutic Hot-Path Proxy (:4000) "]
        Engine["🔒 WASM Policy Engine<br><i>(&lt;5ms Evaluation Latency)</i>"]
        DLP["🔐 Secret DLP & Masking"]
        PCAS["🛡️ PCAS Action Primitives<br><code>BYPASS</code> | <code>ENHANCE</code> | <code>HIJACK</code> | <code>KILL</code>"]
    end

    subgraph SyncDaemon[" 🔄 Sync Daemon "]
        Reconcile["Bidirectional Config Reconciler<br><i>(Harness Config Sync)</i>"]
    end

    subgraph UpstreamProviders[" 🌐 Upstream Providers "]
        Providers["Anthropic API / OpenAI / LiteLLM / Ollama"]
    end

    Agent -->|1. Tool Call / Prompt| HotPathProxy
    SOP -->|2. Rule Sync| Reconcile
    Reconcile -->|3. Hot-Reload Rules| Engine
    HotPathProxy -->|4. Clean Request| UpstreamProviders
    Engine -->|5. Block/Hijack Verdict| Agent

⚡ 30-Second Quickstart

1. Install the CLI & Native Proxy Gateway

# Install global CLI and native Rust proxy binary
npm install -g @intutic/cli @intutic/proxy

# Or run the native proxy directly on-demand
npx @intutic/proxy

2. Connect Your Workspace

Run intutic connect inside your project folder. This boots the local high-speed Rust proxy on port 4000 and auto-detects installed coding assistants:

intutic connect

3. Route Any Agent to Intutic

Point your favorite LLM client or agent framework to the local proxy:

# Anthropic SDK / Claude Code append /v1/messages themselves — host only, no /v1
export ANTHROPIC_BASE_URL="http://localhost:4000"
export OPENAI_BASE_URL="http://localhost:4000/v1"

That's it! Your agent is now governed by real-time safety guardrails.


🔥 Key Features

Feature Description
Sub-5ms WASM Engine Policy evaluation overhead stays under 5ms, preserving typing speed and agent execution fluidity.
🛡️ Zero-Trust Tool Interception Intercepts dangerous commands (rm -rf, git push --force, DROP TABLE) before they touch your system.
🔐 Secret DLP & Masking Automatically redacts API keys ([REDACTED_SECRET]), AWS credentials, and tokens in prompts & tool payloads.
💰 Session Spend Ceilings Prevents "loop burn" by enforcing token spending ceilings per session (e.g. $5.00 limit).
🔄 18 Harness Adapters Auto-detects and config-syncs 18 harnesses — Claude Code CLI, Cursor, Windsurf, Aider, Antigravity, Cline, and more.
🤖 Single & Multi-Agent Swarms Governs single developer tools natively, plus any OpenAI-compatible framework (LangGraph, CrewAI, AutoGen) launched via intutic exec or base-URL env vars.
🎯 Local Model Routing Thompson-sampling bandit picks the best model per task and learns from deterministic local rewards — no LLM judge, no control plane. Enable via intutic_settings.routing. Docs
🧩 Author Your Own WASM Rules Compile AssemblyScript policies with intutic policy compile / install / list-local into ~/.intutic/wasm — the proxy hot-reloads them within ~5s. Ships with an intutic-rule-author agent skill. Docs

🛡️ The 4 PCAS Primitives

Every tool call and prompt evaluated by Intutic produces one of four PCAS Action Primitives:

 ┌──────────┐  ┌───────────┐  ┌────────────┐  ┌──────────┐
 │  BYPASS  │  │  ENHANCE  │  │   HIJACK   │  │   KILL   │
 └────┬─────┘  └─────┬─────┘  └─────┬──────┘  └────┬─────┘
      │              │              │              │
      ▼              ▼              ▼              ▼
 Direct Pass    Inject Safety  Redact Secrets   Hard-Abort
 (<1ms Overhead) Context SOP    or Swap Args     Runaway Loop
  1. BYPASS: Standard safe execution passes through natively (<1ms overhead).
  2. ENHANCE: Inject contextual SOP prompt rules or architectural guidelines.
  3. HIJACK: Substitute dangerous tool parameters or redact secrets on the fly.
  4. KILL: Hard-abort execution thread if an agent attempts destructive file/git ops or hits loop caps.

🔌 Supported Harnesses & Frameworks

Intutic ships 18 harness adapters that are auto-detected and config-synced without modifying your agent's source code. Anything else that speaks an OpenAI- or Anthropic-compatible API is governed the same way by pointing its base URL at the proxy:

Category Supported Tools & Frameworks
Single-Agent Assistants (adapters) Claude Code CLI, Cursor, Windsurf, Aider, Antigravity, Cline, Roo Code, Codex, Continue, Claude Desktop, Goose, Pi, GitHub Copilot, OpenWebUI
Multi-Agent Swarms (adapters) OpenHands, OpenClaw, Hermes, n8n
Any OpenAI-compatible framework (no adapter needed) LangGraph, CrewAI, AutoGen, and anything else honoring OPENAI_BASE_URL / ANTHROPIC_BASE_URL — launch it with intutic exec or export the base-URL env vars

📝 Write Your First SOP

Intutic governance rules are written in standard Markdown files inside your repository root (CLAUDE.md, .cursorrules, or .windsurfrules). Intutic automatically syncs and enforces them in real time:

# Standard Operating Procedure (SOP): Safety Guardrails

## Rules
1. **No Secret Leaks**: Agents must never output raw API keys or passwords.
2. **File Boundaries**: Restrict file modifications to the current project directory.
3. **No Force Push**: Block `git push --force` on all branches.

## Denied Commands
- `rm -rf`
- `DROP TABLE`
- `TRUNCATE`

💬 Interactive Slash Commands

Because Intutic evaluates prompts pre-flight, you can run interactive governance commands directly inside your agent chat:

/intutic status   # View active session spend and compliance score
/intutic rules    # List active WASM & Markdown SOP rules

📚 Documentation & Community


⭐ Star Us On GitHub

If you find Intutic useful, please give us a star on GitHub! It helps us support more agent harnesses and policy engines.

Star on GitHub


🏢 Enterprise & Commercial Licensing

For custom VPC deployments, enterprise-grade SSO/SAML, dedicated SLA support, or team compliance auditing, visit intutic.ai or contact us at [email protected].


📄 License

This project is licensed under the MIT License.

Yorumlar (0)

Sonuc bulunamadi