ADAM (AI)
ADAM is SREInsights’ AI assistant. It works only on an instance where the operator has set ADAM_API_KEY (an Anthropic API key). Without it, ADAM’s features answer that they are not set up, and the rest of SREInsights works as usual.
What ADAM does
Section titled “What ADAM does”| Feature | Where | What it does |
|---|---|---|
| Ask ADAM | Ask ADAM page (/ai) |
Reads a ticket, an alert or a log excerpt you paste, and lists what’s affected, the errors and codes it quotes, the checks SREInsights could run to catch this next time, and next steps. |
| Draft with ADAM | Flow editor toolbar, and Ask ADAM | Drafts a flow from a description, such as “log in to the shop API, list stuck orders, compare the count with the queue depth in Redis”. It uses your connections by name, is checked by the same rules as a save, and lands on the canvas for you to review. |
| Write with ADAM | Next to the code editor of a script check or script step | Writes a Node, Python or shell script from a description, against the runner’s input and output contract, or changes the script in the editor. The code lands in the editor only when you accept it. |
| Export as tests | Flow editor toolbar | Turns the flow on the canvas into a Playwright, pytest, k6 or Postman file to commit and run in CI. Connection settings become environment variables; the drawer lists them, the dependencies, and anything it could not translate. |
| Propose a fix | Under an analysis on Ask ADAM | Paste the function or file involved and get a diagnosis, a unified diff, the updated file, risks, and tests to add. |
| Find the code | Inside Propose a fix | Pick a GitHub or GitLab connection, a repository and a runner group. ADAM reads file names and line numbers from the incident’s stack traces, and a runner in that group fetches the lines around them. The excerpts fill the editor. |
What ADAM is sent, and what it isn’t
Section titled “What ADAM is sent, and what it isn’t”- Secrets are never sent. Connections are named, never opened.
- Only what you paste, or excerpts a runner fetched. SREInsights never connects to your repository itself: Find the code is done by one of your runners, with the connection’s sealed token, over GitHub’s or GitLab’s API. Hosted runners can’t be used for it.
- The audit log records which files were asked for, never their contents.
Limits
Section titled “Limits”- On an instance that runs as AWS Lambda functions, a request has 29 seconds. A slow answer from the model can fail at the gateway; try again.
- ADAM’s drafts are suggestions. A drafted flow is validated like any save, but you review it before it runs.