Open-source, provider-agnostic, security-first AI agent that runs 100% locally. No cloud. No telemetry. No lock-in.
Cloud AI agents send your code, files, and secrets to remote servers. Rivet keeps everything on your machine.
Where it runs
Privacy
Cost
Provider lock-in
Audit trail
Offline mode
A complete AI agent toolkit — from code generation to security sandboxing — all running on your hardware.
Reads files, writes code, runs shell commands, fetches URLs. A general-purpose AI agent for any dev task.
Real-time web UI with approval cards, live streaming output, and full conversation history.
Default-deny permissions, encrypted secrets vault, content guard, and process sandboxing.
Multi-step automations with human approval gates, retry policies, and structured error handling.
OpenAI, Anthropic, Google, Ollama, Groq — or any OpenAI-compatible API. Switch in one line.
Works fully offline with local models via Ollama. No internet required. No data leaves your machine.
Up and running in under 60 seconds.
Five security pillars baked into the architecture, not bolted on as an afterthought.
Every destructive action requires explicit human approval. The agent proposes, you decide.
Encrypted secrets vault with scoped tokens. Each tool gets only the permissions it needs.
User input delimiters and source labeling prevent data from crossing trust boundaries.
Prompt injection detection and data exfiltration prevention built into the agent pipeline.
Domain allowlisting and egress filtering ensure the agent only reaches approved endpoints.
Works with your favorite providers
Rivet is built by PulseSpark AI, an AI consulting firm in Pittsburgh, PA. We build AI solutions for businesses — from strategy and training to production-grade tools like Rivet.
Rivet is MIT-licensed and community-driven. Contributions, issues, and forks are always welcome.