github.com/rebuno/rebuno

Execution runtime for production agents


Keywords
agent-governance, agent-infrastructure, agent-runtime, ai, ai-agents
License
MIT
Install
go get github.com/rebuno/rebuno

Documentation

rebuno

Rebuno is an open-source execution runtime for production agents.

It records every tool call and LLM call as a durable step. Interrupted runs resume from the last recorded step instead of starting over. Any step can be allowed, denied, or held for human approval.

Quick Start

Prerequisites: Go 1.26+, Python 3.11+ / Node 22+

Start the dev kernel:

go run ./cmd/rebuno dev --config examples/rebuno.dev.yaml

Start an agent in another terminal:

Python

pip install rebuno
python examples/python/hello.py

TypeScript

npm install rebuno
npx tsx examples/typescript/hello.ts

Create an execution and follow its event log:

rebuno exec create hello '{"query": "hello world"}'
rebuno exec watch <id>

Documentation

Start here:

Reference:

  • Agents: how an agent process receives work and drives its effects.
  • Tools: effects, step identity, and idempotency.
  • LLM calls: intercepting LLM requests so they replay durably.
  • Streaming: live token deltas while a step is running.
  • Policy: the YAML rule language for allow / deny / require-approval.
  • Events: the event types and their payloads.
  • HTTP API: the /v0 client, agent, and admin endpoints.
  • CLI: the rebuno binary and its commands.
  • Deployment: running the production kernel, config, and Docker.
  • Python SDK: building with Python
  • TypeScript SDK: building with TypeScript
  • Dashboard: web UI to view executions, steps, events, and agent activity.

License

MIT