TSOK · The Source of Knowledge
An AI lab building autonomous organizations humans can trust.[1]
Open research, a governed agent stack, and honest engineering — run as a community, not a company.
[1] Autonomous does not mean uncontrolled. Human oversight must be architected into the system from the ground up. Read the manifesto. — The Manifesto
Oversight by design
Autonomous does not mean uncontrolled. Humans set policy and guardrails; agents execute inside them.
Transparency as foundation
Every agent decision explainable and auditable. Trust is earned through tamper-evident mechanisms, not promises.
Open source as imperative
The infrastructure of autonomous organizations must be inspectable. The defaults here are open.
The stack
Six mechanisms, one thesis
Identity, governance, runtime, fleet control, data protection, isolation — each project contributes a mechanism, and they compose without owning each other.
- IdentityapdShort-lived cryptographic identities for agents
- GatewayMCPGOne governed MCP URL: policy, audit, redaction
- RuntimeagentdA 3 MiB supervised agent, no local code execution
- FleetagentctlKubernetes control plane over an open contract
- DLPSCRUBRReversible secret masking for LLM traffic
- ApplianceAgentic OSThe agent as PID 1, zero egress by construction
Featured
Current focus
Latest
Recent work
Directory
All projects
Honest maturity labels on everything: research, sandbox, graduated.
- MCPGOne governed MCP URL for every backend — identity, policy, audit on every call.
- agentdA 3 MiB static binary that runs one supervised AI agent — MCP-native, no local code execution.
- agentctlA Kubernetes control plane that provisions, scales, and secures fleets of conformant AI agents.
- SCRUBRKeep secrets out of LLM traffic — reversible masking that rehydrates mid-stream.
- mcp2cliTurn any MCP server into a first-class command-line application.
Community
Built in the open. Join in the open.
No hiring pipeline — contribution is the interview. Adopt a sandbox project, file honest friction, bring a workload.