An agentic AI lab. We build infrastructure and systems for the agentic age.

• Axion — Agent cognitive middleware. Inspect, detect, and verify agent reasoning in real time. A proxy layer that makes agent beliefs visible, catches revision loops, and blocks bad actions before they execute.
• AxiContext — Project context controller. Manage and inject structured project context into agent workflows, ensuring agents stay grounded in your codebase and requirements.

• PolyBrain — Multi-model request routing for agentic workflows.
• PolyGnosis — Structured multi-agent debate (planner, proposer, challenger, evidence, judge) for evaluating and converging on stronger LLM outputs.
• PolyMesh — Open-sourced protocol for agent-to-agent communication. Agents discover each other, exchange capabilities, and delegate tasks. Local-first, no internet required. (MIT)
• PolyFlow — Multi-step agent pipelines. Chain agents together: plan, execute, validate.
• PolyScribe — GitHub-native release editor. Draft, edit, and publish release notes directly from your repository.

• LexGateway — SaaS platform for ultra-low-latency LLM access via a global edge network.
• LexRouter — Automatic model selection. Send a task; it picks the cheapest, fastest model that can handle it.
• LexRapid — One proxy, two models. Drafts fast with an 8B model and optionally improves with a larger model. 8B speed, 70B quality. Invite-only.
• LexVerdict — Verification API for agent results. POST a tool call, goal, and result — get back pass or steer with a reason, in milliseconds. Open-source (MIT).
• LexShield — Policy firewall for agent tool calls. Define guardrails that approve, reject, or flag tool invocations before they execute.

• VisCompile — Behavioral diff tool for AI agents. Snapshot golden test suites, diff after prompt changes, catch regressions before they ship. OSS CLI (Rust) + hosted SaaS.
• VisReplay — Session recorder and debugger for AI agents. Record every decision, replay step-by-step, fork at any point and re-run. OSS SDK (TypeScript) + hosted SaaS.

• VekTor — Task-oriented agents through a simple API. Describe what you need; VekTor fetches, manipulates, and acts.
• VekData — "Give me the Q3 revenue by region" -> agent queries your database, builds the table, returns it.
• VekInbox — Durable human approval queue. Route agent actions to humans for approval, with persistent queues and async review workflows.

• VisCompile — Behavioral diff tool. Compare agent runs side-by-side to surface regressions and behavioral changes.
• VisReplay — Session recorder and debugger. Capture full agent traces and replay them step by step for debugging.
• VisBoard — Shared agent-human workspace. Collaborative dashboard for monitoring and interacting with running agents.

• ForgeDistill — Correctness-by-construction distillation for agentic tool-calling models. Deterministic tool-call chains with real dependencies, teacher writes prose only. Training data that is structurally perfect and grounded.

Make agentic systems more reliable by letting specialised agents reason together, critique each other, and converge on better answers - without hiding the humans in the loop.

• Website: latticeag.vercel.app
• Articles: Dev.to
• GitHub: @LatticeAG