Start here.
Agent Forge is a private preview — there is no public account creation, so these docs are for people who already have access, or are deciding whether to request it.
Access is granted by hand.
There's no signup form to fill in. Send a request, describe the workflow you want to run, and it's reviewed directly — usually within a few days.
Request a demo →The four ideas worth understanding first.
The evidence layer that scores every model on capability, health, latency, and quality history — and what it uses that score for.
Read on Platform TrimurtiThe three-reviewer council that checks correctness, policy, and cost before higher-risk work leaves the system, and what crosses that threshold.
Read on Platform MemoryWhat's persisted between requests — sessions, projects, checkpoints, and vector retrieval — and where each layer actually lives.
Read on Platform WorkflowsWhat a request actually looks like end to end: routing decision, subtask fan-out, action preview, approval, trace, rating.
Read WorkflowsAPI
The platform is operated through the browser interface, not a public API — this is a private preview with owner-only access. One reference endpoint is documented for integrators already granted access:
- A single
/searchendpoint over indexed content, authenticated the same way as the rest of the platform - Not a general-purpose platform API — routing, agents, and Data Grid are not exposed this way today
What's shipped recently.
Agent Forge ships continuously during private preview. A curated changelog for preview users is in progress — until it's live, ask directly and you'll get a straight answer about what's changed since you last used it.