M

MeshAI

On the RadarAI Compliance IndexPaidAI Governance & Risk Management

OpenTelemetry-native "agent control plane" from MeshAI Labs (founder ex-AWS): auto-registers every AI agent that emits telemetry, attributes token spend by team and agent, enforces policy via OTLP ingest or an inline LLM proxy, and generates EU AI Act Article 12 records and Article 73 incident evidence from runtime traces. Pre-GA pilot programme as of August 2026.

Last verified

Visit Site GitHub

on the radar · vendor-claimed

This is an early-stage listing. All information below is vendor-claimed and has not yet been independently verified by AI Compliance Index. No compliance score is issued and this tool is excluded from rankings and comparisons.

A radar listing is not an endorsement. These are early-stage companies, some pre-launch, and everything here comes from the vendor. Do your own due diligence before you buy, sign a contract, or send them any data. Read our full disclaimer.

embed this badge

<a href="https://aicompliance.ai/tools/meshai"><img src="https://aicompliance.ai/badges/on-the-radar.svg" alt="On the Radar — AI Compliance Index" width="260" height="64"></a>

Free to embed. The badge must link to your listing and may not be altered.

overview

MeshAI positions itself one layer above conventional observability. Datadog, New Relic, Honeycomb, Grafana Tempo and Jaeger collect agent telemetry; MeshAI ingests the same OpenTelemetry gen_ai.* traces and interprets them against a governance policy, producing the artifacts an APM cannot: Annex III risk classification, Article 6(3) derogation documentation, Article 12 record-keeping, Article 26 deployer-accountability bundles, and Article 73 incident reporting. It deploys in parallel with an existing stack through one YAML change in the OpenTelemetry Collector rather than replacing anything.

Four capability pillars: an auto-populating agent registry that flags unregistered and shadow agents; ML anomaly detection running four algorithms on a five-minute cadence for cost spikes, reliability decay, model-swap drift and dormant-agent reactivation; token-level cost attribution with budget guardrails and forecasting; and policy enforcement with eight policy types, human-in-the-loop approval queues, an agent kill switch, ABAC, and prompt-injection and PII detection at the proxy. Framework coverage spans CrewAI, LangGraph, AutoGen, LlamaIndex, Pydantic AI, Semantic Kernel, Copilot Studio and Salesforce Einstein, plus coding agents via an MCP server.

Pricing is public and unusually transparent for this category: a free forever tier for one agent, then $299/mo (25 agents), $799/mo (100 agents), $1,999/mo (1,000 agents), and custom Enterprise Plus. EU-region single-tenant and customer-hosted deployment are available on request. MeshAI states it never stores prompt or completion content — only operational metadata.

Maturity signals, stated plainly: MeshAI Labs is early. It is recruiting 5-10 enterprise pilot partners ahead of GA, the GitHub organisation holds five small repositories with negligible community traction, and SOC 2 is described as roadmap rather than achieved. Its stated conformance target, prEN 18229-1 under CEN/CENELEC JTC 21, is a January 2026 working draft, so it is a target and not a certification. The company contributes to the OpenTelemetry semantic-conventions working group (issues #159 and #160).

One credit where due: MeshAI's central technical claim is anchored to a real, checkable source. It says it implements the feedback edge from post-market drift detection back to risk-management reassessment in the twelve-step compliance architecture proposed in "AI Agents Under EU Law" (Nannini et al., arXiv:2604.04604, 6 April 2026). We verified that the paper exists, has those authors, and does propose a twelve-step architecture.

core features

  • MeshAI positions itself one layer above conventional observability.
  • Datadog, New Relic, Honeycomb, Grafana Tempo and Jaeger collect agent telemetry; MeshAI ingests the same OpenTelemetry gen_ai.* traces and interprets them against a governance policy, producing the artifacts an APM cannot: Annex III risk classification, Article 6(3) derogation documentation, Article 12 record-keeping, Article 26 deployer-accountability bundles, and Article 73 incident reporting.
  • It deploys in parallel with an existing stack through one YAML change in the OpenTelemetry Collector rather than replacing anything.
  • Four capability pillars: an auto-populating agent registry that flags unregistered and shadow agents; ML anomaly detection running four algorithms on a five-minute cadence for cost spikes, reliability decay, model-swap drift and dormant-agent reactivation; token-level cost attribution with budget guardrails and forecasting; and policy enforcement with eight policy types, human-in-the-loop approval queues, an agent kill switch, ABAC, and prompt-injection and PII detection at the proxy.
  • Framework coverage spans CrewAI, LangGraph, AutoGen, LlamaIndex, Pydantic AI, Semantic Kernel, Copilot Studio and Salesforce Einstein, plus coding agents via an MCP server.

target frameworks

EU AI Act

faq

MeshAI — frequently asked questions

What is MeshAI?
MeshAI positions itself one layer above conventional observability. Datadog, New Relic, Honeycomb, Grafana Tempo and Jaeger collect agent telemetry; MeshAI ingests the same OpenTelemetry gen_ai.* traces and interprets them against a governance policy, producing the artifacts an APM cannot: Annex III risk classification, Article 6(3) derogation documentation, Article 12 record-keeping, Article 26 deployer-accountability bundles, and Article 73 incident reporting. It deploys in parallel with an existing stack through one YAML change in the OpenTelemetry Collector rather than replacing anything.
Which compliance frameworks does MeshAI support?
MeshAI addresses EU AI Act. Check the vendor site for the latest scope and any region-specific certifications.
How much does MeshAI cost?
MeshAI is a paid product. Visit https://meshai.dev for the current pricing tiers.
Who is MeshAI for?
Teams working on ai governance & risk management — particularly those that need to evidence EU AI Act controls.
Is MeshAI open source?
Yes — the source code is available on GitHub at https://github.com/meshailabs-org.