Getting Started
Quick Start
Prerequisites: Go 1.25+, Python 3.10+
1. Start the kernel
go install ./cmd/rebuno
rebuno dev
You should see:
rebuno dev — development mode
kernel http://127.0.0.1:8080
policy permissive (all tools allowed)
storage in-memory (data lost on restart)
Waiting for agents...
2. Start the hello world agent
pip install rebuno
python examples/agent/hello.py
The agent connects to the kernel via SSE and waits for executions.
3. Create an execution
rebuno create --agent hello --input '{"query": "hello world"}'
Or with curl:
curl -s -X POST http://localhost:8080/v0/executions \
-H "Content-Type: application/json" \
-d '{"agent_id": "hello", "input": {"query": "hello world"}}' | jq
The kernel assigns the execution to the agent. The agent proposes two tool calls (reverse and word_count) as intents, the kernel approves each one, and the agent returns the result:
{
"query": "hello world",
"reversed": "dlrow olleh",
"word_count": 2
}
4. View the event log
Every action is recorded as an immutable event:
rebuno events {id}
Or with curl:
curl -s http://localhost:8080/v0/executions/{id}/events | jq
Replace {id} with the execution ID from step 3. You'll see events like execution.created, intent.accepted, step.created, step.completed, and execution.completed — a full audit trail of what the agent did and what the kernel decided.
You can also follow events in real time with rebuno events --tail {id}.
5. Add a policy
Restart the kernel with a policy to see what happens when tools are denied:
rebuno dev --policy examples/policies/hello.yaml
Now reverse and word_count are allowed, but any other tool will be denied. To see a denial, you can modify the hello agent to call an unlisted tool — the kernel will reject the intent and the agent will receive a PolicyError.
Add the Explorer
Requires: Node.js 18+
To also start the web-based execution viewer:
cd explorer
npm install
npm run dev
The explorer is available at http://localhost:3000. It connects to the kernel at http://localhost:8080 by default. It shows executions, event timelines, and step details.
Next Steps
- Architecture — Core concepts, workflow, and state transitions
- Deployment — Production setup, authentication, and configuration reference
- Python SDK — Building agents and runners
- Policy — Declarative policy rules
- CLI — CLI reference