Build one agent. Give it every channel.

Quickstart

Create an organization and an agent in the Keiki dashboard. Open API Keys, issue a key with the agents:chat scope, and enable the API gateway on the agent you want to call. The agent name or ID becomes the model value in provider-compatible requests.

The OpenAI-compatible base URL is /api/v1/openai. Existing OpenAI clients can list enabled agents as models and start a persisted agent turn through chat completions.

curl https://onkeiki.com/api/v1/openai/chat/completions \
  -H "Authorization: Bearer $KEIKI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"support","messages":[{"role":"user","content":"Where is my order?"}]}'

Authentication and keys

Developer APIs use bearer API keys. Create and revoke keys from the dashboard, bind a key to one agent when an integration should not reach the rest of the organization, and grant only the scopes that integration needs. Never place a secret key in browser code or a public repository.

OpenAI-compatible requests accept Authorization: Bearer. Anthropic-compatible requests also accept x-api-key. MCP uses OAuth discovery and authorization metadata published by this deployment.

Sandbox and test workflow

Use the dashboard playground as the test environment before connecting a customer-facing channel. It runs the configured agent with its real instructions and tools while keeping the conversation separate from production contacts. Eval Studio can replay saved cases and compare model or prompt changes before release.

For API integration tests, create a key bound to a non-production agent, keep its gateway enabled only while testing, and use the x-conversation-id response header to continue or isolate a conversation deterministically.

Interfaces

The OpenAPI document describes the public HTTP contract with typed request and response schemas. The MCP endpoint uses Streamable HTTP so compatible agent clients can discover and call Keiki tools. AgentKit supports both cloud-served agents, synced to Keiki, and self-hosted agents that run the same declarative configuration on your infrastructure.

• OpenAI-compatible models and chat completions
• Anthropic-compatible messages and token counting
• Model Context Protocol over Streamable HTTP
• AgentKit configuration sync and self-hosted serving
• Conversation traces, evaluations, and scoped API keys

Versioning and deprecation

Keiki versions its public REST API in the URL path, beginning with /api/v1. Breaking contract changes use a new major path rather than silently changing existing integrations.

When an operation is scheduled for removal, Keiki marks it deprecated in OpenAPI and returns Deprecation and Sunset response headers with migration guidance before the removal date.