Agent on Demand
Ravi internal cloud Agent infrastructure API
Agent on Demand

Ship agent workflows without building the runtime.

Give your product a clean API for long-running coding agents: isolated sandboxes, versioned configs, live event streams, resumable sessions, and reliable teardown.

Sandboxed sessions Live event streams Versioned configs Bring your own keys

Session primitive

From user action to running agent in three calls.

  1. POST /agents Package the model, runtime, tools, and instructions.
  2. POST /sessions Launch the job from your app with prompt and resources.
  3. GET /sessions/<id>/stream Stream progress, logs, status, and output back to users.

Reusable workers

Agents

Define the agent once with model, runtime, instructions, MCP servers, skills, and default resources.

Configurable

Live work units

Sessions

Run work in isolated sandboxes, resume later turns, terminate jobs, and keep the full event trail.

Streamed

Production context

Environments

Attach env vars, packages, repositories, setup scripts, and network rules without leaking them into UI code.

Controlled

Product patterns

Patterns

Power review bots, research fleets, triage queues, CI agents, CLI wrappers, and internal ops tools.

Shippable

You build the product experience

Quickstart

AOD runs the machinery

Deploy guide

Small integration surface

Turn any product action into agent work.

When a user clicks run, review, research, triage, or fix, your app creates a session and streams the result back. AOD stays behind the scenes as the agent control plane.

# Start a run
curl -X POST https://aod.ravi.id/sessions \
  -H "Authorization: Bearer $AOD_KEY" \
  -H "Content-Type: application/json" \
  -d '{"agent_id":"<id>","prompt":"review this PR"}'

# Stream the work back into your UI
curl -N https://aod.ravi.id/sessions/<id>/stream \
  -H "Authorization: Bearer $AOD_KEY"

Runtime adapters

claude codex gemini opencode your runtime next