Everything Anlyon does, on one page
Your agent acts in your users' accounts. Here is every part that keeps that safe: what it can call, whose account it uses, what is checked before anything leaves, what is recorded after, and the tools you build it with.
A few lines on your server, then connect a user.
Your agent names an action. Anlyon makes the call.
- ActionsDeclare an HTTP call once. Your agent invokes it by name and never holds the key.
- IntegrationsEach integration as a full toolset: every read, and governed writes.
- PacksInstall an integration's actions ready-made. Plan the install, read it, then apply it.
- OpenAPI importTurn a spec into actions, with risk, approval and impact suggested for each operation.
- Any HTTPS APIDeclare what a call changes and how to read it back, and it is governed like the rest.
- Hosted functionsYour own JavaScript or Python, with no network and no keys. Its writes are planned and approved.
- SecretsA vault with no read endpoint. Each secret is bound to the hosts it may be sent to.
- Your own vaultPoint a secret at AWS Secrets Manager or HashiCorp Vault. Anlyon reads it at the call.
In each user's own account. Under that user's rules.
- Act for a userYour server names the user with Anlyon-End-User. The model's tools have no user field.
- Connect linksA link per user and integration: a short consent page, a key form, or straight to the provider.
- Your own OAuth appSet your client id per environment, and the consent screen shows your app's name.
- A limit for each userAn impact limit scoped to each end user: every user's agent gets the whole amount.
- Approvals your users giveA policy decided by the end user sends the approval to them, inside your app.
- React componentsApproval and connect screens for your app, on a token that lasts minutes.
- Erase a userOne call deletes their connections, resets their limits and cancels their approvals.
Before the call. What is checked, and what can stop it.
- Approval policiesRules on every call, with conditions, a version history and policy files you can test.
- Bound approvals and previewsThe yes is for the exact request. A changed action or amount is refused, not sent.
- Team approvals and SlackCalls that are yours to decide reach your team, in the console or with Slack buttons.
- Anlyon VigilAn opt-in second reader for your approval queue, with a suggestion on each request.
- Impact limitsOne allowance for what agents change, in money or your own unit, reserved before the call.
- BudgetsA monthly cap per API key, with an alert before it is reached.
- EnvironmentsA key belongs to one environment, so staging can't reach production.
- Kill switchOne switch halts every call in an environment, approved ones included.
- Protection and promotionProduction changes on purpose, and a rollback is a pointer move.
After the call. What is recorded, and who hears about it.
- Receipts and gradesEvery call's record, graded by how far its outcome was checked.
- Read-backsConfirmed means a read of the provider's own record matched the write.
- Idempotency and replayA retry with the same key is answered from the record, never run twice.
- Unknown outcomesA lost response is recorded as unknown, held, and never sent again.
- Audit recordEach change is written in the same transaction as the change itself.
- Audit streaming and exportStream to a signed webhook, S3, Datadog or Splunk, or export a date range.
- Events and webhooksA signed POST when an approval is asked or decided, a call ends, or an environment halts.
- Notifications and alertsIn the console, by email, to a webhook, or to Slack or Discord.
Your team. Who can do what, in each environment.
- Roles per environmentAdmin, member, approver and viewer. A person's role can differ per environment.
- Single sign-onOIDC or SAML for each workspace, required if you choose.
- SCIM directory syncYour directory adds and removes people, and revokes the API keys they created.
- Emergency accessOne or two admins keep a password sign-in, so a directory mistake can't lock you out.
- Workspace defaultsWho decides a user's call, how long a request waits, and who hears when something goes wrong.
Single sign-on, SCIM, audit streaming, your own vault and roles per environment depend on your plan.
See pricingBuild with your stack. And your agents' clients.
- TypeScript SDK@anlyonhq/sdk, with forUser to bind a client to one user.
- Python SDKanlyon on PyPI, with client.for_user(id).
- Vercel AI SDK toolsYour actions as AI SDK tools, with a connect link when a user hasn't connected.
- MCPA hosted MCP server for Claude Code, Cursor and Codex, and one per integration.
- CLIInvoke actions, decide approvals and script CI from a terminal.
- Framework examplesCrewAI, Mastra, Pydantic AI, the OpenAI Agents SDK and the AI SDK, wired to actions.