Cross-framework orchestration for AI agents

Mix LangGraph, CrewAI, LangChain, AutoGen, and custom agents without race conditions. ProtoMesh provides a shared state fabric, conflict resolution, and policy-based governance so teams can ship production-grade multi-agent systems.

Policy-based governance
Cross-framework state fabric
Optimistic + consensus locks
orchestratorRaft candidate · 4 nodes
agent
LangGraph
agent
CrewAI
policy
budget<=$100
fs
sandbox:/workspace
net
egress: allow-list
rag
source: approved
trace
[00:01] intent: write(/workspace/report.md)
[00:01] lock: entity(report.md) → granted
[00:02] policy: budget ✓ fs ✓ net ✓
[00:03] write: 1.2KB → ok
[00:03] unlock: entity(report.md) → ok

Works with

LangGraph
CrewAI
LangChain
AutoGen
LlamaIndex
OpenAI Swarm
Feature

Shared state fabric

Agnostic state store with idempotency keys, optimistic concurrency, and per-entity locks so agents never step on each other.

Feature

Governance as code

Pre-execution hooks and allow-lists across tools, RAG, network, and FS. Enforce budgets, RBAC, and audit trails by default.

Feature

Mix any framework

Adapters for LangGraph, CrewAI, LangChain, AutoGen, and custom agents. Keep your stack; standardize the coordination layer.

How it works

  • Agents send intents via a typed envelope.
  • ProtoMesh mediates access to shared state with idempotent, conflict-aware ops.
  • Governance hooks evaluate policies before tool calls, RAG, network, or FS I/O.
  • Emit traces for full cross-agent observability.

Why use ProtoMesh

  • Avoid framework lock-in while using the best tool for each agent.
  • Prevent race conditions and data corruption, not just observe them.
  • Enforce compliance, budgets, and RBAC from day one.
  • OSS-first, cloud optional. Deploy anywhere.

Governance as a first-class primitive

Define deterministic policies that gate every critical action. Evaluate in-process for latency, with remote signing for auditability.

Tool call allow/deny
RAG source allow-list
Network egress rules
FS path sandboxing
PII redaction
Budget & rate limits

MVP 0.2v scope

  • • LangGraph + custom agents
  • • Optimistic locking + per-entity mutex
  • • Basic policies: budgets, rate limits, allow-lists
  • • Traces + minimal dashboard

Get early access

We're onboarding a small group of design partners. Tell us about your stack and use-case, and we'll reach out.

Or email service@protomesh.dev