kamayoq
An open-source control plane for agentic operations. It runs AI agents under policy, sends each tool call through one gateway, and keeps an audit record of each action.
Status: Pre-alpha. Under design. Not ready for use.
Components today
kamayoq
Control plane
Defines agents, runs them, and applies policy and approvals.
tambo
MCP gateway
Checks scopes, injects secrets, and applies rate limits for each agent tool call.
quipu
Audit store
Writes a hash-chained, append-only record of prompts, tool calls, decisions, and results.
Planned
These parts are on the roadmap. They do not exist yet.
- Permission-aware knowledge search
- Evals
- Workflow templates
Why it exists
Every team that puts agents into production builds the same pieces: a gateway for tool calls, a policy check, an approval step, and an audit log. Most teams build them late, after an incident. kamayoq puts those pieces in one place, as open source, so they come first.
What it is not
- Not an agent framework. kamayoq runs agents on adapters, with LangGraph as the default.
- Not a model provider or an LLM proxy. It calls providers through the LiteLLM SDK.
- Not a multi-tenant SaaS. It is single-tenant and self-hosted.
How it relates to my work
kamayoq is where I write down the design I use in engagements. You do not have to adopt kamayoq to work with me. When it is ready for production use, clients can choose it. Until then, I build the same controls with tools that are ready today.
Follow the work
The code and the design issues are public on GitHub. Issues and pull requests are welcome.
Read the code and the issues on GitHub. License: Apache-2.0.
Want these controls without the wait?
Engagements put the same design in place with tools that are ready today. Book a call to talk about your workflow.
Book a 30-minute call