Roadmap

AgentWatch Roadmap

AgentWatch is currently executing the v0.4.x milestone, building out the foundational components of the V2 Invariant Lattice Architecture following our massive Brutalist Purge.

🎯 Current Focus (v0.4.x)

• The Invariant Lattice: Implementation of the three core execution stages: CapabilityLattice, StateLattice, and CognitiveLattice.
• Shadow Simulators: Building out the ShadowDatabase, ShadowFilesystem, and ShadowNetwork modules to allow the StateLattice to safely test agent actions against mock environments before execution.
• InvariantEvolver: Implementing statistical analysis of execution histories to automatically suggest relaxations or tightenings of safety thresholds based on false-positive patterns.
• Core Stability: Ensuring the EventBus and async processing pipelines can handle high-throughput telemetry without race conditions or memory leaks.

❌ Out of Scope (Do Not Submit PRs For These)

As part of our commitment to YAGNI (You Aren't Gonna Need It), the following domains are strictly out-of-scope for the foreseeable future:

1. Reasoning Auditors: We will not accept PRs attempting to audit LLM internal reasoning states or fingerprinting logic.
2. Complex Governance Systems: EU AI Act compliance trackers, heavy SQL logging schemas for unneeded auditing, and regulatory reporting are dead.
3. Speculative Abstractions: "Prepared" hooks, custom client-side exponential backoff math, and overly abstract generic wrappers will be rejected.

🚀 Long-term Vision (v1.0)

• Complete coverage of system-level Shadow Simulators for all standard agent tools.
• True plug-and-play local agent tracking via lightweight adapters.
• Seamless, zero-config visual dashboard out-of-the-box.
• Rock-solid stability with maximum 0.1ms latency impact on agent operations.