PyClaw¶
Last updated: 2026-06-28
PyClaw is a local-first agent orchestration framework that can use Cortensor for scalable or verifiable inference.
Product Role¶
PyClaw is an agent runtime/client, not part of the base protocol. Its job is to run local agent loops, tools, memory, policy, and channels while routing selected work to Cortensor when remote execution or validation is useful.
Module Map¶
| Module family | Role |
|---|---|
pyclaw-runtime |
Turn orchestrator and daemon host. |
pyclaw-brain |
Agent loop for planning, acting, observing, verifying, and responding. |
pyclaw-session |
Append-only event log for messages, tool calls, approvals, verdicts, and summaries. |
pyclaw-context |
Prompt packing and context budget management. |
pyclaw-llm |
Provider abstraction with an integration path for Cortensor router routing. |
pyclaw-tools |
Tool host and plugin system. |
pyclaw-policy |
Approval and grant system. |
pyclaw-memory |
Long-term and working memory. |
pyclaw-skill |
Versioned procedural playbooks. |
pyclaw-a2a |
Local agent-to-agent delegation queue. |
Additional Modules¶
PyClaw also includes supporting modules that make the local runtime auditable and extensible:
| Module family | Role |
|---|---|
pyclaw-storage |
SQLite-backed storage for sessions, memory, skills, registry, and policy. |
pyclaw-artifact |
Content-addressed file store for logs, downloaded files, generated files, and large artifacts. |
pyclaw-compress |
Rolling summaries and long-session compaction. |
pyclaw-identity |
Versioned agent profile capsule. |
pyclaw-metagovernor |
Optional meta-control for risk, uncertainty, budget, and validation escalation. |
Agent Loop¶
PyClaw's agent loop is:
interpret -> plan -> policy-gate -> act -> observe -> verify -> reflect -> respond
That makes PyClaw a natural client for Cortensor validation: a local agent can delegate heavy execution to the router, then ask Corgent/router validation routes whether an output is safe enough to use.
Cortensor Relationship¶
flowchart LR
User["User / channel"] --> PyClaw["PyClaw local runtime"]
PyClaw --> Tools["Local tools and artifacts"]
PyClaw --> Router["Cortensor router"]
Router --> Miners["Miner sessions"]
Router --> Validation["PoI / PoUW / consensus"]
Validation --> PyClaw
Product Status¶
PyClaw is positioned as a product-facing agent framework when its Cortensor integration path is active. The product surface focuses on local orchestration, policy, tools, memory, sessions, and selected Cortensor-backed execution or validation.