Introducing Yaha Agents

Most teams do not need another chatbot demo. They need an assistant with a known prompt, an allowed model, tools they chose, and a version they can point production at.

Yaha Agents are workspace-scoped assistants you build in the console. You edit a draft, publish an immutable version, test it in a playground, then run it from the UI, a deployment, or a workflow.

This is not the same as pointing Cursor or Claude Code at the Gateway. Those are coding tools that call Yaha. A Yaha Agent is an assistant you define.

Configure once, publish when ready

The builder covers the decisions that usually live in a wiki and a half-finished script:

• Identity — persona, goals, and system prompt
• Intelligence — model plus Chat, ReAct, Reflection, Plan & Execute, or Auto
• Capabilities — skills, tools, and MCP
• Outputs — markdown, JSON, table, CSV, or a schema
• Safety — internet, tools, moderation, and PII controls

Save the draft as you iterate. Publish pins the production snapshot. Later edits go to a new draft until you publish again.

Test with traces, not guesswork

The builder Playground is a chat against that version. Runs show status, tokens, cost, and step traces. If a run should not proceed without a person, you can require human approval.

Create agents from the list, a template, or an Ask Yaha proposal.

Deploy after publish

A published version can go to:

• REST API (deployment key)
• Website widget
• Slack, Teams, or Discord
• MCP
• A workflow step

External callers use the deployment key on the public agent chat endpoint. Project yaha.* keys stay on the Gateway for OpenAI-compatible apps.

Getting started

Open Products → Agents in the console, create an agent, publish a version, and send a playground message. Full steps live in the Agents docs.

Ready to try it? Sign in and publish your first agent.