Agents

Sandbox an agent in under two minutes: identity, budget, bound key, demo turn, then scale the fleet.

Autonomous agents need a scoped credential, a budget, and clean attribution — not a shared workspace key. In Relixr an agent is a first-class identity with a bound rlx_ key.

0→1 sandbox

1. Open Agents → Start agent sandbox (or pick a template: Support, Code review, Research).
2. Defaults: $5 daily cap, frontier model allowlist, development environment.
3. Copy the key once from the First run panel (or run the in-dashboard demo after topping up any amount).
4. Confirm the turn under Logs — usage is tagged with agentId.

$ curl https://api.relixr.com/v1/chat/completions \
-H "Authorization: Bearer $RELIXR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"claude-fable-5","messages":[{"role":"user","content":"agent sandbox ok"}]}'

Agent groups

Groups share spend limits and allowlists while each agent keeps its own key. Prefer setting the budget on the group once, then attaching agents.

Management API sketch

GET    /api/agents/groups
POST   /api/agents/groups
PATCH  /api/agents/groups/{id}
DELETE /api/agents/groups/{id}

Point any OpenAI-compatible agent runtime at the bound key — no Relixr-specific protocol required.