Skip to content
Skip to main content
Edwards Tech Innovation
Edwards Tech InnovationTechnology that takes care of you.
All Labs projects
BetaFor: AI teams, platform engineers, and developers building multi-agent workflows.

CAM-PROTOCOL

Complete Arbitration Mesh

Open-core governance and routing for AI agent workflows, with a policy-enforcing MCP gateway.

Get notifiedGitHub repositoryDocumentation (in repo)
npm install @cam-protocol/complete-arbitration-mesh

Status: Beta

The open-core software is released at v2.1.2 with a passing 468-test suite. The managed Pro cloud tier is still in development and not yet buyable.

  • 468 tests passing (4 skipped, 0 failing)

The problem

Why this exists

  • Teams run several AI providers at once, each with different capabilities, prices, and failure modes — and no consistent way to choose between them per request.

  • Agent workflows call external tools with no policy layer: no way to say "never send PII to this tool" or "this tenant gets 100 requests per minute."

  • Multi-agent setups are hard to coordinate — discovering agents, splitting tasks, and auditing what each agent did.

  • Compliance teams need audit trails and cost attribution across AI usage, which raw provider APIs do not provide.

The approach

How it works

CAM sits above your providers and MCP servers as an arbitration and governance layer. A FastPath routing engine picks the best provider for each request based on cost, latency, and policy.

The MCP gateway enforces policies, trust tiers, rate limits, and audit logging on every tool call — governance that applies to any MCP-compatible workflow.

Inter-agent collaboration handles agent discovery, task decomposition, role-based collaboration, and secure messaging, with OpenTelemetry instrumentation throughout.

Capabilities

Key features

FastPath routing

Route each request to the optimal AI provider by cost, latency, and policy — with automatic failover.

MCP governance gateway

Policy-based tool selection, trust tiers, data-classification filtering, and per-tenant rate limiting above any MCP server.

Audit logging

Every decision and tool call logged with trace IDs for compliance and debugging.

SDK integrations

Official SDKs for OpenAI, Anthropic, Google, and Azure, plus streaming responses and response caching.

Inter-agent collaboration

Agent discovery, task decomposition, role-based collaboration, and secure inter-agent messaging.

Observability

OpenTelemetry instrumentation, health checks, retry logic, and detailed performance metrics.

Under the hood

Architecture overview

CAM Core — routing engine

Request arbitration across providers: cost/latency scoring, failover, response caching, and rate limiting.

MCP Gateway

Governance layer above MCP servers: policy enforcement, trust tiers, data-classification filters, audit logging.

Inter-Agent Collaboration (IACP)

Agent discovery, task decomposition, role assignment, and secure messaging between specialized agents.

Shared infrastructure

Authentication and authorization, configuration, provider and MCP connectors, metrics and telemetry, and a dedicated security layer.

Terms

Licensing & pricing

Licensing

  • Community edition: Apache License 2.0 — free for any use.
  • Pro and Enterprise editions: commercial license (license key or agreement required).

Pricing

  • FreeCommunityApache-2.0, works out of the box
  • $99/moProPer workspace — planned, in development
  • CustomEnterpriseBy inquiry — planned

Pricing above is the published plan from the CAM editions. The managed Pro tier is not yet live — talk to Dru about early access.

What's next

Roadmap

  • Managed Pro cloud tier with a hosted, multi-tenant MCP gateway.

  • Expanded provider integrations and policy templates.

  • Hosted evaluation and cost-benchmarking dashboards.

Interested in CAM-PROTOCOL?

The managed Pro tier is in development. Get notified when it opens, or talk to Dru about a pilot.

Get notified