Multi-Agent Awareness & Outer Loop: OMNI v0.5.7

We are thrilled to release OMNI v0.5.7, a major milestone that brings agents out of their silos. This release introduces Multi-Agent Awareness, persistent memory, and an automated outer loop for pattern discovery.

Multi-Agent Awareness & Persistent Knowledge

Agents shouldn’t have to relearn your project every time they start a session.

• omni_agents: A new MCP tool allowing agents (like Claude, Cursor, and Copilot) to discover and interact with each other’s state on the same project.
• omni_knowledge: Cross-session memory for agents to permanently learn and store project-specific quirks and filter preferences.

Advanced ROI & Meta-Harness Outer Loop

We’ve integrated powerful diagnostics directly into the agent toolkit:

• omni_history & omni_budget: Direct MCP access to distillation logs and ROI simulators, empowering agents to track their own token efficiency.
• omni optimize: Our new Meta-Harness Outer Loop automatically validates generated LLM filters, closing the loop on autonomous pattern discovery.

Distiller Enhancements

• Non-Bash Tool Routing: The engine now handles ReadFile, Grep, and WebFetch output seamlessly.
• Context Preservation: Build and Test distillers now preserve --> contextual error blocks.
• Positional Scorer Boost: Dynamic priority bumping for active errors in multi-line outputs ensures critical signals are never dropped.