Know what your agents did. Prove it later.
axonpush shows the business work your agents complete, what waits and on whom, and seals every event into tamper-evident evidence your auditors and enterprise customers can check.
Your coding agent drafts the workspace over MCP. You review it before anything is live.
- Claude Code
- Codex
- Cursor
- Windsurf
- LangChain
- LangGraph
- CrewAI
- OpenAI Agents
- Anthropic SDK
- Vercel AI SDK
- Mastra
- LlamaIndex
- Google ADK
- OpenTelemetry
Every agent run returned fine. The invoice is still waiting on a supplier, nobody knows who acts next, and when an auditor asks what the agent decided last March, the answer is spread across five tools and a log that rotated weeks ago.
From agent run to evidence
- 01 · Observe
The work shows up in your words.
Your coding agent reads the app and drafts the entities, events and views to track. Once you activate it, the queue fills: what needs attention, what waits on whom, and what is in flight.
- 02 · Explain
Every case carries its evidence.
Open any item for its linked records and timeline. The traces sit underneath, down to the inputs of the tool call that failed, and one click hands the evidence to your coding agent.
- 03 · Seal
Every hour is sealed and locked.
Events are written to encrypted hourly segments, Merkle rooted, chained to the hour before and stored under Object Lock. A daily RFC 3161 timestamp means history cannot be quietly rewritten, even by us.
- 04 · Share
Hand over proof, not screenshots.
Build a signed evidence pack for a period and the frameworks you are asked about. Give auditors a scoped link that expires, and let anyone check it with the open verifier.
Evidence for the clauses auditors check
Enterprise buyers and regulators ask for proof of what your agents did and who oversaw them. axonpush keeps it, maps it to the clauses they cite and shows the gaps that are still yours to close.
- EU AI Act
- ISO/IEC 42001
- GDPR
- DORA
- NYC LL144
- AIUC-1
- OWASP Agentic Top 10
- NIST AI RMF
- Colorado SB 26-189
- CCPA ADMT
- 01
Tamper-evident by construction
Hourly segments with per-event Merkle proofs, a chained seal for every hour, daily RFC 3161 timestamps and Object Lock storage. Verify the whole chain, or prove a single event, from the dashboard or offline.
- 02
Privacy that survives an audit
Content capture you control, per-subject keys that make erasure final, legal holds that pause expiry, and retention tiers that keep a year of history without a year of hot storage.
- 03
Oversight with a paper trail
A register of your agents, their owners and grants. Human reviews, denials and notices are recorded as evidence, and incident clocks count down to each reporting deadline.
axonpush produces evidence. It is not legal advice or a certification, and it never marks you as compliant. Mappings to rules that are not yet in force are flagged for review.
Set up where the code gets written
Paste one prompt into your coding agent. It connects over MCP, reads your app, drafts the workspace and adds the observe calls. You review a plain-language change list before anything is live.
- Claude Code
- Codex
- Cursor
- Windsurf
- + any MCP client
- Python
- TypeScript
- .NET
- OpenTelemetry
- LangChain
- LangGraph
- CrewAI
- OpenAI Agents
- Anthropic SDK
- Vercel AI SDK
- Mastra
- LlamaIndex
- Google ADK
How youe sees every agent in Nova Room
Nova Room is a hiring marketplace where candidate and company agents meet, and people approve the steps that matter. youe runs it on self-hosted axonpush to see what every agent is doing, while candidates' data stays scoped, short-lived and erasable.
- Every agent followed from first connection to first action
- Sessions replayed next to their traces
- Scoped keys, 30-day content and erasure that holds
Pay for what your agents send
Start free. SSO, RBAC and the audit log come with every plan, and Team adds the evidence you hand to auditors.
Free
Try it on one real workflow.
$0/month
- 250,000 events/month
- 7 days recent history
- 30 days archive
- 3 seats
- SSO, RBAC and the audit log
- Ingestion pauses at 110% until next month.
Pro
For a product running agents in production.
$49/month
or $490/yr
- 2,000,000 events/month
- 14 days recent history
- 180 days archive
- Unlimited seats
- SSO, RBAC and the audit log
- Over the limit? Buy a top-up of 500,000 events for $30.
Get startedUpgrade any time in Settings → Billing
Team
The evidence plan, for teams that answer to auditors.
$299/month
or $2,990/yr
- 4,000,000 events/month
- 30 days recent history
- 1 year archive
- Unlimited seats
- SSO, RBAC and the audit log
- Evidence packs
- Auditor links
- Public trust page
- Compliance lock
- Legal holds
- Over the limit? Buy a top-up of 2,000,000 events for $100.
Get startedUpgrade any time in Settings → Billing
Scale
Custom volume and long archives, set up with you.
From$1,500/month
- Custom event volume
- 30 days recent history
- Up to 7 years archive
- Unlimited seats
- SSO, RBAC and the audit log
- Evidence packs
- Auditor links
- Public trust page
- Compliance lock
- Legal holds
- Volume is agreed with you.
Self-hosted or in your own cloud? Talk to us.
Over your limit, Pro and Team can buy a top-up. Stored data is never deleted early.
Questions, answered
Who is axonpush for?
Teams running customer workflows across agent runs and human handoffs, especially AI companies that sell to enterprises and get asked for proof of what their agents did.
Can I sign up today?
Yes. The hosted cloud is open. Create an account, copy the setup prompt into your coding agent and review what it drafts.
Does axonpush make us compliant?
No tool can. axonpush records evidence, maps it to the clauses of frameworks like the EU AI Act, ISO/IEC 42001 and DORA, and shows the gaps that are still actions on your side. It is not legal advice or a certification.
How is the record tamper-evident?
Each hour of events is written as an encrypted segment with a Merkle root, chained to the previous hour's seal and stored under Object Lock. The chain is timestamped daily by an RFC 3161 authority, and any event can be proven with a downloadable proof.
What data does axonpush store?
Only the keys your workspace declares. Undeclared keys are dropped and counted. You choose whether prompt and response content is stored in full, redacted or not at all, and fields marked personal are masked for most readers.
How does erasure work with an immutable archive?
Personal content is encrypted with per-subject keys held outside the archive. Erasing a subject destroys their keys, so sealed history stays verifiable while their content can no longer be read.
Can auditors see the evidence without an account?
Yes. Signed evidence packs can be shared through scoped links that expire, and every access is logged. The open verifier checks signatures and proofs without contacting axonpush.
What counts as an event?
Each span, log, error or observation your app sends. A batch counts once per item in it.
What happens when I reach my limit?
On Free, ingestion pauses at 110% until next month or until you upgrade. Pro and Team can buy a top-up. Stored data stays either way.
Do I have to replace my tracing?
No. axonpush has SDKs for Python, TypeScript and .NET, and it ingests OpenTelemetry spans over OTLP. Traces stay attached to the items they explain.
Start with one prompt.
Create an account, then give this prompt to your coding agent. It connects, reads your app and drafts what to track for you to review.
Set up axonpush for this repository. axonpush adapts per installation: you decide, with me, what this app tracks, and I review it before it goes live. I should not need to copy any keys or ids from the dashboard; the MCP tools give you everything.
API: https://api.axonpush.xyz
Dashboard: https://app.axonpush.xyz
Docs: https://docs.axonpush.xyz/agent-operations
Use our own deployment URLs if these differ.
1. Connect the axonpush MCP server at https://api.axonpush.xyz/mcp if it is not connected yet (for Claude Code: `claude mcp add --transport http axonpush https://api.axonpush.xyz/mcp`). Sign-in happens in the browser; there is no token to paste. Call the workspaces_describe tool to learn the spec format, the roles, the draft ops and the setup steps. Do not guess tool names or fields; use what describe returns.
2. Use the workspace I gave you. If I did not give you one, list the workspaces and use one that is not active yet, or create one for our app with a blank spec ({"name": "..."}).
3. Read this codebase. Find the things worth tracking (for example agents, tickets, orders, runs), the events that change them, and the attributes each one needs. Prefer opaque ids and bounded labels. Leave out prompts, responses, message bodies, documents, secrets and raw URLs.
4. Propose a data dictionary, entities and events to me in plain language before writing anything. For every attribute you want to mark personal (names, emails, companies), say why it is needed. Only declare personal attributes I agree to.
5. Read the shared draft with workspaces_draft and change it with workspaces_applyDraftOps, sending the draft version each time. Make small, typed edits; never rewrite the whole spec. Check workspaces_draftChanges and fix any issues it reports.
6. Ask me to review and activate the draft in the dashboard (https://app.axonpush.xyz). If I approve it here in the chat, call workspaces_activateDraft yourself; otherwise wait for me.
7. Call workspaces_connect for the workspace. It mints a key that can only publish, bound to this app and one environment, and returns an env map. Write that env into the app's git-ignored env file (.env.local, or the project's existing secret convention) and check the file is ignored. Never commit, print or paste the key. If connect says a key already exists and this repository does not have it, call it again with rotate: true.
8. Add the axonpush SDK to the app and record events with the SDK's observe call, at the points you identified. Each event carries schema_version, source_event_id (stable across retries), event, occurred_at, an optional environment, refs (entity type to opaque id), optional trace_id and span_id, and attributes limited to the keys you declared. Use the SDK's identify call to attach the declared profile attributes (such as a display name) to an entity. Read the configuration from the env written in step 7. Follow the docs at https://docs.axonpush.xyz over anything in this prompt. Keep telemetry out of business transactions so a failure here never changes app behaviour.
9. Send one test event from the app and confirm it appears in workspaces_catalog.
Finish with the diff, what each event means, which attributes are personal, where you wrote the env, and anything you could not verify.