# Avowgate

> Avowgate keeps a checklist of what your AI coding agent promised, ties each item to a test, and blocks done until the tests pass.

Avowgate is a hosted service at https://avowgate.com/. It is for people who ship software with AI coding agents and need to know what to trust.

## How it works

1. Say what you want. Before coding starts, the agent writes your request down in plain words.
2. Make promises. Your request becomes specific statements that either hold or do not.
3. Link the tests. Each promise points at the tests that prove it. A promise that nothing tests shows as open.
4. Pass the gate. The work cannot be closed until every promise has passing proof.

## Connect an AI agent

Create an access token under Settings in the app (https://avowgate.com/app/), then add an MCP server over HTTP:

```
claude mcp add --transport http avowgate https://avowgate.com/api/v1/mcp --header "Authorization: Bearer YOUR_TOKEN"
```

Address: https://avowgate.com/api/v1/mcp
Header: Authorization: Bearer YOUR_TOKEN

Order of calls:

1. `at_list_products`: find the product this project belongs to, or create it with at_create_product.
2. `at_task_start`: record what the user asked for before writing code.
3. `at_promise`: write one testable promise for each behavior the user asked for.
4. `at_create_test`: register each test by its exact name, then link it to its promise with at_link_spec_test.
5. `at_report_results`: report the real test results for the commit, using the short commit id.
6. `at_check_gate`: ask for the verdict before telling the user anything is done.
7. `at_task_close`: close the task; it only succeeds when the gate passes.
8. `at_record_deployment`: record which commit is running where, after each deployment.

An agent cannot sign off a risk, dismiss a finding, promote a gate rule or retire a test. A person does those.
More for agents: [For AI agents](https://avowgate.com/agents/)

## Questions and answers

### What is Avowgate?

Avowgate is a hosted service that keeps a checklist of what your AI coding agent promised to do, links each item to a test, and refuses to call the work done until those tests pass. The answer is a gate that reads pass or blocked, with the reasons.

### How is that different from asking my AI agent whether it is done?

An agent summary is a claim. Avowgate replaces it with a verdict built from test results tied to a specific commit. A promise with no test shows as open, and the gate stays blocked until every promise has passing proof or a named person accepts the risk.

### How do I connect my AI coding agent to Avowgate?

Create an access token under Settings in the app, then add Avowgate as an MCP server over HTTP. In Claude Code the command is claude mcp add --transport http avowgate https://avowgate.com/api/v1/mcp --header "Authorization: Bearer YOUR_TOKEN". Any client that supports MCP over HTTP can use the same address and header. Nothing needs to be installed.

### Which AI coding agents work with Avowgate?

Claude Code and any other agent or tool that can talk to an MCP server over HTTP. There is also a command line tool and hooks for setups that do not use MCP.

### What does the gate block, and when does it pass?

The gate blocks closing a task while any promise is untested, failing, waiting for a person to sign off, or covered only by a test that has not run on the current commit. It passes when every promise has passing proof for the commit, or an accepted risk recorded by a person.

### What can an AI agent not do in Avowgate?

An agent can read everything, write requirements, link tests, and report results and findings. It cannot sign off a risk, dismiss a finding, promote a gate rule or retire a test. Those decisions belong to a person and are recorded with their name.

### What data does Avowgate store, and does it use my content for AI?

It stores the requests, requirements, test names, test results, commit identifiers, deployment records and notes that you or your agent send. It does not run your tests or fetch your code. It does not send your content to AI services and does not use it to train AI models.

### What kinds of projects is Avowgate for?

Any software project where an AI agent writes code and a person needs to know what to trust. Starter packs add the checks that are easy to forget: web security, payments, accounts and passwords, and safe delivery.

### Can I try Avowgate before signing up?

Yes. The home page has a read-only live demo that needs no account. Sign-up is currently by early access: leave your email on the home page and you will be written to when there is room.

### How do I reach the people behind Avowgate?

Use the contact form at https://avowgate.com/contact/ or write to info@avowgate.com.

## Pages

- [Home](https://avowgate.com/): What Avowgate is and a read-only live demo
- [For AI agents](https://avowgate.com/agents/): How an agent connects and the order of calls
- [Guide](https://avowgate.com/docs/): Requirements, tests, the gate, sign-offs, findings, CI
- [Contact](https://avowgate.com/contact/): Questions and privacy requests
- [Privacy](https://avowgate.com/privacy/): What is collected and how long it is kept
- [Terms](https://avowgate.com/terms/): Terms of service
- [Service providers](https://avowgate.com/subprocessors/): Who handles data for the service