Kimss — Secure AI Gateway for Autonomous Agents

Deprecated for new gateway onboarding. Do not pip install kimss to connect an app. Keep the native OpenAI or Anthropic client and follow the control-plane agent-to-agent contract. New Python agents: Kimss Forge (pip install kimss-forge, then gateway="kimss"). This package remains for existing control-plane callers and the optional IDE MCP server.

Track, govern, and secure autonomous agents with exactly 1 line of code. Zero data-plane refactoring required.

Kimss is a Model-Agnostic Enterprise AI Gateway and Governance Control Plane with a dual-listener inbound proxy (OpenAI + Anthropic). Keep the native SDK you already use — swap the base URL, add an Agent ID header, and every call gets identity, an audit trail, and a kill switch. Kimss never hosts your models (BYOI).

AI coding assistants: read the canonical control-plane AI_INTEGRATION.md first (local mirror: AI_INTEGRATION.md).

OpenAI

from openai import OpenAI

client = OpenAI(
    base_url="https://api.kimss.ai/v1",
    api_key="kimss_workspace_key",
)
response = client.chat.completions.create(
    model="custom:kimss-gpt-5-3",
    messages=[{"role": "user", "content": "Execute audit."}],
    extra_headers={"X-Kimss-Agent-Id": "enterprise_auditor"},
)

Anthropic

from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.kimss.ai",
    api_key="kimss_workspace_key",
)
response = client.messages.create(
    model="custom:kimss-claude-3-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Execute audit."}],
    extra_headers={"X-Kimss-Agent-Id": "enterprise_auditor"},
)

Or zero-code:

OPENAI_BASE_URL="https://api.kimss.ai/v1"
OPENAI_API_KEY="kimss_..."
# or
ANTHROPIC_BASE_URL="https://api.kimss.ai"
ANTHROPIC_API_KEY="kimss_..."

Start with a 14-day Production trial (no card): 100,000 governed requests/month during the trial · Sign up

flowchart LR
  App[Your_app_or_agent] --> GW["Kimss_Gateway"]
  GW --> Model[Vaulted_provider]
  GW --> Trail[Governed_audit_trail]

3-step setup

1. Sign In & Vault

Log into Kimss AI. Open Governance → Connected Infrastructure / Provider Vault and vault your model provider endpoint + key.

2. Mint Key

On the Gateway tab, Generate Key. Copy the kimss_... Control-Plane workspace key once. Same keys under Governance → API Keys.

3. Route Traffic (zero refactoring)

Point your OpenAI client at https://api.kimss.ai/v1 or your Anthropic client at https://api.kimss.ai, set the key, and add X-Kimss-Agent-Id (see hero snippets).

Step-by-step: GETTING_STARTED.md · 5-minute tutorial: kimss-ai/kimss-python-quickstart · control-plane OpenAPI: kimss-ai/kimss-control-plane.

Control plane (DevOps) — optional pip install kimss

The kimss package is an infrastructure management client — not an inference SDK. Prefer the OpenAI gateway for all chat/completions.

pip install kimss

Inference wrappers (agents.run, chat, models.create, …) are deprecated — see CHANGELOG.md.

Optional: MCP for IDEs

Cursor / Windsurf / Claude Desktop can run kimss-mcp-server (pip install 'kimss[mcp]' or uvx --from kimss[mcp] kimss-mcp-server). MCP inference tools are deprecated; use the OpenAI gateway from application code. Plugin layout: .cursor-plugin/plugin.json.

Examples

See examples/00_gateway_proxy.py (OpenAI) and examples/00b_anthropic_proxy.py (Anthropic).

License

MIT — see LICENSE.