Developer quickstart

Run a governed agent from Python

The published kimss package wraps supported agent, model, file, and vector-store paths. Optional MCP exposes a focused tool subset for IDE workflows.

Last updated: July 23, 2026

Install and authenticate

Install the published kimss package, store KIMSS_API_KEY outside source control, and call APIs with X-Kimss-Key or bearer tokens plus X-Workspace-ID when one key spans workspaces.

pip install kimss pulls the supported client. Never commit workspace keys; use secret stores in CI and production.

Use the supported agent surface

Agent run and create, model completions, and file upload form the core SDK—prefer /v1/agents/run for new agent execution paths.

Conversation continuation uses conversation_id in the SDK. MCP exposes run agent, create agent, complete, and file upload for IDE workflows.

Conversation state

Use conversation_id when continuing threads across runs.

Files

Upload ephemeral attachments with client.files.upload. Kimss does not host RAG; retrieve on your side.

MCP tools

Optional MCP server for focused tool access in development environments.

Example flow

Configure credentials, create or select an agent, run with messages, and inspect usage impact on workspace governed-request usage—repeat in staging before production keys.

Full code samples and edge cases live in /docs/python_sdk_mcp. Public REST shapes are documented at /docs/api_docs, not production Swagger.

Keep the boundary explicit

The SDK does not wrap every billing, admin, SCIM, or backoffice route—verify shipped documentation before assuming a client method exists.

Migrate legacy assistant helpers via /assistants-to-v1-migration. Enterprise MCP policy: /mcp-enterprise-ai.