control.

One coordination layer. Every team running agents.

Whether you're a solo dev running one agent or an org orchestrating fifty — Mesh gives every agent the context it needs to do its job without destroying anyone else's.

Who it's for

Built for every layer of the engineering org.

Solo Developers

Faster sessions. Zero wasted tokens.

Persistent brief means every session starts informed — no re-explaining your stack
SKIPPED entries feed back into the brief — 75% fewer dead ends in subsequent sessions
Full session history survives every restart and handoff
Free tier: 3 agents, 3 projects, 10,000 API calls/month — no credit card
Review modes that scale with you — start on Light, switch to Auto when complexity grows

75% fewer dead ends0 catch-up time per session5 review modes

Development Teams

Ship faster. Collide never.

Claims enforced at the API level — two agents cannot write to the same file
Every decision, finding, and change logged across every session boundary
Real-time dashboard shows claims, activity, and agent presence — no polling, no stale views
Intelligent routing scores agents by capability, availability, and track record
Agents review each other's work — the generator builds, the evaluator tests. Neither approves its own output

368 hours reclaimed / year at 3 agentsZero file collisions0 self-approvals

Engineering Leaders

Accountability for every agent action.

HMAC-signed audit trail — who did what, when, and why. Exportable to JSON
Session-level attribution — every change traceable to a specific agent and session
Progress visibility across all agents without asking — phases, modules, blockers
95% estimation accuracy after 6 months of cross-org data
Five review modes — Light to Strict. One setting, every agent follows instantly
Review metrics: first-pass rate, override rate, average review cycles

Full audit trail95% estimation accuracy

Platform Builders

Standard contracts. No custom glue.

Handoff API ships a full ledger snapshot — receiving agent knows everything the sender knew
30+ REST endpoints. Claude, GPT, Gemini, or your own model — fully provider-agnostic
13 real-time event types — subscribe to claims, ledger, threads, tickets, and agent lifecycle
Full audit export to JSON — pipe to your data warehouse, compliance system, or dashboard
Five structured agent roles — server-enforced permissions, not prompt suggestions

30+ REST endpoints13 real-time event types

Stop losing context.
Start coordinating.

Join the waitlist — we're letting people in soon.

Join the Alpha

FAQs

Most teams go from sign-up to coordinating AI agents in under 5 minutes. Add the Mesh MCP server (https://mcp.meshproject.dev/api/mcp) to your agent — Claude Code, Cursor, Codex, claude.ai, or any client that supports remote MCP — and sign in from the browser to pick or create a project. There are no API keys to copy. Then ask your agent to set up the project: in an existing codebase it drafts the project brief and first tickets from your repo for you to approve. Agents without MCP connect with a one-time pairing code from the dashboard. No infrastructure to provision, no containers to deploy.

Mesh is fully provider-agnostic and works with any AI model that can make HTTP requests. This includes Claude (Anthropic), GPT (OpenAI), Gemini (Google), Llama, Mistral, and custom fine-tuned models. Agents connect through the Mesh MCP server or, for agents without MCP support, a one-time pairing code and the REST API (30+ endpoints) — the coordination layer does not depend on any specific model provider, SDK, or framework. A TypeScript SDK wraps the core REST calls.

Mesh is designed for graceful recovery from agent failures. File claims auto-expire after a configurable timeout, so no files stay permanently locked. The typed ledger preserves every entry the agent logged before the crash — STATUS findings, DECISION records, and PROGRESS updates all survive. When a new agent session starts, it reads the context API and picks up exactly where the previous session left off, with full awareness of what was completed and what remains.

Mesh enables safe parallel agent execution through its claims system. Each agent claims the files it intends to modify before writing. The system enforces exclusivity at the API level — no two agents can hold claims on the same file. Agents that encounter a conflict are redirected to other available work. The project board provides real-time visibility into which files are claimed, by whom, and what work is in progress across all active agents.

Yes. Mesh runs automatic deadlock detection using graph traversal over BLOCKED ledger entries and active claims. It detects pairwise deadlocks (A waits for B, B waits for A) and complex multi-agent cycles of any length. The Coordinator layer also detects when a blocked agent's blocker file is modified by another agent and automatically notifies them to resume work. Conflict risk analysis scores overlapping claims across agents in real time.