Platform

The capability list, with a status against every row

Feature grids in this category tend to blur what ships with what is planned. Everything below carries a status, and "in development" means exactly that — not "ask your rep".

Coverage

26 runtimes, one control plane

Your engineers did not standardise on one agent and they are not going to. ClawMetry treats the runtime as a detail: the same cost, activity, and policy surface works across all of them.

Coding agents

Claude Code, Codex, Cursor, Copilot, Gemini CLI, Cline, OpenHands, Aider, Goose, OpenCode, Devin, Antigravity, Kimi, Grok, Qwen Code, DeepSeek Harness and more.

Agent frameworks and harnesses

OpenClaw, NVIDIA NemoClaw, Hermes, DeepAgents, n8n, Exo, Pi, and self-hosted harnesses. New runtimes are added on request as part of an enterprise engagement.

01 · See

Visibility

Cost attributionSpend by agent, session, model, repository, tool, and person — derived at ingest so aggregates stay consistent.Shipping

Live activityA streaming view of what agents are doing right now, including tool calls and reasoning steps.Shipping

Waste and anomaly detectionRunaway loops, repeated context, and cost anomalies flagged against your own baseline.Shipping

Session replayFull transcripts and tool timelines for any session, for incident review.Shipping

Fleet viewEvery machine running agents, with per-node health and cost.Shipping

OpenTelemetryMetrics, traces, and logs ingested over OTLP; export to your own collector.Shipping

02 · Control

Governance

Off unless you turn it on. A tool that silently changes how your agents behave is not something a platform team can safely deploy.

Budget limitsPer-node and per-team spend ceilings, with alerts before the ceiling and enforcement at it.Shipping

Tool policyCall-level risk scoring with policies that block or require approval by risk level.Shipping

Approval queueRisky actions held for a human decision, routed to the reviewer, with approve-and-remember.Shipping

Kill switchStop a running agent, or a fleet of them, from the dashboard.Shipping

Policy guardrailsNVIDIA NeMo Guardrails integration for policy enforcement layered on any runtime.Shipping

AlertingCustom rules with webhook delivery to Slack, PagerDuty, Opsgenie, or your own endpoint.Shipping

Scope note

Enforcement depth varies by runtime. Some agents expose a pre-tool hook we can gate on; others only expose telemetry after the fact, so for those we can observe and alert but not block. We will tell you exactly which of your runtimes fall into which category before you buy — ask for the conformance matrix.

03 · Prove

Evidence and audit

Audit logEvery agent action recorded with a hash chain, so tampering is detectable rather than merely discouraged.Shipping

Secret redactionEvents are scrubbed before they are stored or forwarded anywhere, including to your SIEM.Shipping

SIEM exportSyslog over RFC 5424 in CEF or JSON — Splunk, QRadar, ArcSight, Elastic, or any compliant receiver. Runs on your side, where the plaintext is.Shipping

Compliance evidence bundlesGenerated evidence mapped to NIST AI RMF and SOC 2 control families, with a three-state status per control and gaps reported as gaps.Shipping

RetentionConfigurable, and unlimited on enterprise terms.Shipping

Enterprise access control

Where we are not finished yet

This is the section most vendors would blur. Two of these rows are the reason some deals should wait a quarter, and you should know that before the first call rather than after the pilot.

Single sign-onOIDC-native SSO is designed and in development. SAML is supported through a self-hostable bridge when a contract requires it.In development

Role-based access controlAn owner / admin / member / viewer ladder is in development. Today the model is owner plus member with a full/limited access flag.In development

Air-gapped licensingOffline licence keys verified locally against a signed public key, with no callback.Shipping

Data residencySelf-hosted and air-gapped give you complete control. Managed cloud runs in a single region today; another region is a contract conversation, not a switch.On request

SCIM provisioningNot built. If your process requires it, say so on the first call.Not available

Why we publish this

An overclaim discovered during your security review does not cost us a feature request, it costs us the deal and costs you a quarter. We would rather lose the ones we are not ready for and keep the ones we are.

Get the written detail before you spend a meeting on it

The security whitepaper covers architecture and data flows. The procurement pack covers everything a vendor-review process asks for.

See the documents Talk to sales