Build, orchestrate, and deploy AI agents with guardrails, memory, and accountability. Supports Claude, OpenAI, Gemini, Grok, Mistral, and local models.
npx nestor-sh install
AI agents are powerful but unpredictable. They hallucinate, overspend, execute unsafe commands, and operate as opaque black boxes with no accountability trail.
Nestor gives agents guardrails, memory, and accountability. Every action is sandboxed, every token is tracked, and every decision has a trust score.
A complete platform from CLI to dashboard, built for security and reliability.
Claude, OpenAI, Gemini, Grok, Mistral, and Ollama. Unified API with automatic retries, model routing, and cost optimization across all providers.
Preview every action before it executes. See the full plan, estimated cost, and potential side effects. Approve, modify, or cancel before anything runs.
Every agent gets a grade from A to F based on accuracy, safety, cost efficiency, and reliability. Built from real execution history, not benchmarks.
Agents understand your entire codebase through retrieval-augmented generation. Semantic search across files, functions, and documentation.
Orchestrate multi-agent teams with directed acyclic graphs. Parallel execution, governance gates, cron scheduling, and built-in templates.
Localhost-only by default — no login needed on your machine (like VS Code or Jupyter). For teams, enable multi-user auth with scrypt, CSRF, rate limiting, and RBAC. Rust security core, Docker sandbox, SSRF protection, secret redaction.
From zero to autonomous agent in under 5 minutes.
Interactive wizard configures your API keys and preferred model provider.
$ npx nestor-sh install
Launch the interactive shell. Your agent has access to tools, memory, and your codebase.
$ nestor shell
Start the web dashboard to monitor agents, edit workflows, and track costs in real time.
$ nestor start
A layered architecture with security at every level.
┌────────────────────────────────────────────────────────────┐ │ nestor CLI │ │ install shell start agent skill workflow test │ └──────────────────────────┬─────────────────────────────────┘ ┌──────────────────────────┴─────────────────────────────────┐ │ @nestor/server (Express + WS) │ │ Auth | Rate-limit | Studio Dashboard │ └────┬──────────┬──────────────┬──────────────┬──────────────┘ ┌────┴────┐ ┌───┴────┐ ┌─────┴─────┐ ┌─────┴──────┐ │ @nestor │ │@nestor │ │ @nestor │ │ @nestor │ │ agent │ │ orch │ │ mcp │ │ db │ │ Runtime │ │DAG exec│ │ 15 tools │ │ SQLite │ │ 6 LLMs │ │ Cron │ │4 resources│ │ 14 tables │ └────┬────┘ └───┬────┘ └─────┬─────┘ └────────────┘ ┌────┴──────────┴──────────────┴─────────────────────────────┐ │ nestor-core (Rust → N-API) │ │ SSRF | Path traversal | Secret redaction | Crypto │ └──────────────────────────┬─────────────────────────────────┘ ┌──────────────────────────┴─────────────────────────────────┐ │ Docker Sandbox (optional) │ │ cap-drop ALL | read-only FS | network=none │ └────────────────────────────────────────────────────────────┘
An honest comparison with other agent frameworks.
| Feature | Nestor | OpenClaw | Paperclip | Cognetivy |
|---|---|---|---|---|
| Multi-LLM Support | 6 providers | 3 | 2 | 4 |
| Dry-Run Mode | Yes | No | No | Partial |
| Trust Score (A-F) | Yes | No | No | No |
| Rust Security Core | Yes | No | No | No |
| Docker Sandbox | Yes | Partial | No | Yes |
| DAG Workflows | Yes | No | Yes | Partial |
| RAG / Codebase | Yes | Yes | No | Yes |
| Web Studio Dashboard | Yes | No | Yes | No |
| Skill Testing Framework | Yes | No | No | No |
| Open Source (MIT) | Yes | Yes | No | AGPL |
Three commands from zero to your first agent conversation.
# Install Nestor and configure your LLM provider $ npx nestor-sh install # Create an agent with Claude $ nestor agent create --name coder --adapter claude --model claude-sonnet-4-6 # Start chatting $ nestor shell Nestor Shell v2.0.0 Agent: coder (claude-sonnet-4-6) Trust Score: B+ | Budget: $5.00/session > Review the auth module for security issues Analyzing src/auth/ ... Found 3 files, 847 lines of code. ## Security Review: Auth Module 1. **Password hashing uses bcrypt** - Good 2. **Missing rate limiting on login** - High severity 3. **JWT secret from env variable** - OK but consider vault Suggested fixes applied in dry-run mode. Run `nestor approve` to apply changes.