Skip to content
@WatchLLM

WatchLLM

Runtime governance and reliability infrastructure for autonomous software systems.



Agents are probabilistic. Infrastructure cannot be.

WatchLLM builds deterministic, local-first infrastructure that governs how autonomous coding agents read, write, and execute code — before unsafe changes reach disk.

The Stack

Layer Component Description
Governance kernel Deterministic local runtime enforcement engine (Python, Tree-sitter AST)
Governance runtime High-performance Rust + Wasm implementation with JSON schema parity
Orchestration orchestrator Policy graph, model routing, and sub-agent spawning
Contracts schemas Central event, rule, policy, and violation schema contracts (v1)
Memory klyd Persistent architectural memory — extract, store, enforce design decisions
Observability replay Execution graph replay, DAG rendering, and runtime forensics
Reliability reliability Adversarial stress-testing, prompt injection, and sandboxed evaluation
Quality benchmarks Ecosystem-wide performance and regression benchmarking
Editor vscode VS Code extension — save interception, diagnostics, and inline governance
Docs docs Architecture, RFCs, threat model, roadmap, and gap analysis
Demos examples Curated scenarios: secrets, boundary enforcement, auth flow

Quick Links

Popular repositories Loading

  1. .github .github Public

    Organization profile and community health files for WatchLLM

  2. docs docs Public

    Central documentation — RFCs, architecture, specifications, and roadmap

  3. watchllm-oss watchllm-oss Public

    WatchLLM open-core: deterministic governance kernel, schemas, VS Code extension, and examples. MIT licensed.

    Python

Repositories

Showing 3 of 3 repositories

Top languages

Loading…

Most used topics

Loading…