No AI in yourcompany acts untilWaxell says yes.
A paste into ChatGPT, a Copilot agent's next step, a coding agent's push, an MCP tool call, a workflow you wrote. Each one gets a yes, a no, or a wait for a person before it happens, and then a line in a record your auditor can read.
30 minutes. No commitment.
SOC 2 Type II attested · ISO/IEC 27001:2022 certified · HIPAA with a BAA · aligned to the EU AI Act and NIST AI RMF
AI you built.
AI you bought.
AI nobody approved.
Workflows your engineers wrote. Copilots and agents you pay for. The ChatGPT tab someone opened with a personal login. All three act on your behalf, with your data, on your bill. Most companies can name the first and none of the third.
Waxell is where you find all three and decide what each may do.
Find the AI agents in your company, including the ones installed last Tuesday.
An agent on each enrolled Mac and Windows device sees the AI tools, extensions and personal accounts in use. Connectors pull in the agents hosted on Copilot, Bedrock, Vertex and Agentforce. Each one gets an owner, a scope and a policy.
Endpoints · browser extension · 20 platform connectors
Inside Discover
Two lines of code show you every call, every token and the bill, per agent.
Add waxell.init() and a run lands as a waterfall: each model call and tool call with its tokens, its cost in dollars across every provider, and the policy check that ran at the end. Open last Tuesday's run and read it step by step, then replay a recorded model call against a new prompt version and compare before it ships.
160+ AI libraries (Python) · OpenTelemetry underneath
Inside Observe
The policy answers before the call goes out: allow, hold, redact or block.
Waxell reads the agent, the tool, the data class, the person the agent acts for and the budget left. A hold goes to a person in Slack with the full context, and their answer is written next to the agent's request.
50+ policy categories · approvals in Slack and the workspace
Inside Enforce
Run the agents that touch money or customers with a check before every step.
Two decorators put a workflow under Runtime. Every step is checked first. A budget that runs out stops the run and records why. The kill switch stops the agent from running, immediately.
wax push · checkpoints and resume · 28 agents, 89 versions in the example tenant
Inside Run
The audit trail is already written. Pick a framework and download it.
Every decision carries the policy that made it. One click produces a signed evidence pack mapped to the EU AI Act, NIST AI RMF, ISO/IEC 42001, OWASP LLM Top 10, HIPAA, GDPR or SOX. Or give your auditor a read-only seat and let them look.
evidence packs · read-only auditor seats
Inside Prove
- Security teamsThe AI your employees installed.ChatGPT tabs, coding agents, extensions, Copilot agents a department built. See all of it, set policy per tool, prove it.Open
- MSPs and resellersThe agents across every client you serve.Deploy governance to every client from one console. Register the deal, keep the margin.Open
- Developers and platform teamsThe agents you ship.Two lines to instrument. Traces, cost, and a policy engine that can say no.Open
Everything on this page is checkable. Here is the list.
- 01
Two lines of code, an import and waxell.init(), instrument an agent. Nothing else in the repo changes.
- 02
Waxell Observe recognizes 160+ Python AI libraries automatically, including LangGraph, CrewAI, OpenAI Agents, Anthropic, Bedrock, Vertex, Mistral and Ollama.
- 03
Every decision is one of four words: allow, hold, redact, block. Each is written down with the policy that made it and the reason.
- 04
A hold goes to a person in Slack or the workspace with the agent, the action and the context. Their answer lands in the same record.
- 05
More than 50 policy categories are enforced while the agent acts: PII, data egress, spend, tool scope, time of day, content safety, human approval and more.
- 06
A kill switch stops the agent from running immediately, with no redeploy and no code change.
- 07
Waxell Endpoints runs on macOS and Windows and deploys through the MDM or RMM you already use. An app scanner inventories AI apps, and an optional browser extension checks what is pasted into AI apps in the page.
- 08
Agents hosted on Microsoft 365 Copilot, Copilot Studio, Azure AI Foundry, AWS Bedrock AgentCore, Google Vertex and Salesforce Agentforce are discovered through their own APIs.
- 09
One URL fronts the MCP tools your agents call. Upstream credentials stay inside it. A tool whose schema changes is held until a person looks.
- 10
Records are kept for as long as you set, up to seven years, with legal hold. Regulated tenants can run on their own Postgres, Redis and object store.
- 11
One click produces a signed evidence pack mapped to the EU AI Act, NIST AI RMF, ISO/IEC 42001, OWASP LLM Top 10, HIPAA, GDPR or SOX.
- 12
Waxell is SOC 2 Type II attested and ISO/IEC 27001:2022 certified, supports HIPAA with a Business Associate Agreement, and is aligned to the EU AI Act and NIST AI RMF. The reports are in the Trust Center. Read it.
Works inside the stack you already use
The record you just built
Now you're
in the path.
Every decision on this page was written down as it happened. That's the whole product, for every kind of AI in the building. Free for your first agents and devices, live in about thirty seconds.
30 minutes. No commitment.
SOC 2 Type II attested · ISO/IEC 27001:2022 certified · HIPAA with a BAA · aligned to the EU AI Act and NIST AI RMF
- discovered Copilot Studio · HR assistantendpoints · acme-corp · new agent, no owner yethold
- llm.call · claude-opus-4 · $0.0182support-agent · run 8f3c1a · traced, priced, within budgetallow
- fetch 10.0.4.22:5432cursor · mpatel · egress outside allowlistblock
- step 14 · budget reached · run stoppedruntime · porsche_911_recall_watch · budget cap · reason on the recordblock
- EU AI Act Art. 14 evidence exportedcompliance · Q4 pack · 412 decisions, 0 gaps, signedallow
Policy, identity, reason, outcome. Exportable. Signed. Yours.
Deciding, before an agent acts, what it is allowed to do, then enforcing that decision and keeping the record. Observability tells you what happened. Governance decides what happens next.