Agents
Agents Overview
What a Yaha Agent is, how it differs from coding-agent integrations, and where to start.
A Yaha Agent is a versioned assistant you configure in the console under Products → Agents. You set identity, model, planning mode, tools, outputs, and safety, then publish a version you can test, run, and deploy.
Yaha Agents run through the Gateway product: model routing, allowlists, and usage tracking apply the same way they do for your apps.
Yaha Agents vs coding agents
| Yaha Agents | Coding agent integrations | |
|---|---|---|
| Where you build them | Console Products → Agents | Cursor, Claude Code, Codex CLI, and similar tools |
| What they are | Assistants you define, version, and deploy | External tools that call the Gateway API |
| Auth | Console session, or a deployment key after you publish | Project yaha.* key on the Gateway |
| Guide | This section | Integrations |
If you want to point Cursor or Claude Code at Yaha, use the coding agent guides. If you want to build an assistant your team owns, start here.
What you can configure
- Identity — name, persona, goals, and system prompt
- Intelligence — model plus a planning mode: Chat, ReAct, Reflection, Plan & Execute, or Auto
- Capabilities — skills, tools / MCP, and memory settings
- Inputs and outputs — typed variables and markdown, JSON, table, CSV, or schema outputs
- Safety — internet, tools, moderation, and PII controls
- Playground — chat-style tests against the draft or published version
Lifecycle
- Create an agent in a workspace (from the list, a template, or Ask Yaha).
- Edit the draft and Save draft.
- Publish to pin a production version.
- Run it from the builder playground, the Runs page, a workflow, or a deployment.
Next steps
- Create and run an agent
- Deployments
- Workflows — call an agent as a step in a DAG