Coming soon
What A22A does not do yet, and what to do in the meantime.
These are not built yet. Each one lists what works today.
Listing runs and projects
There is no endpoint to list runs or projects. A run can only be retrieved by its run_id, so keep the id that runs.invoke returns. A project can be retrieved by name.
Waiting for a run
The library has no runs.wait. Until it does, receive the result on your webhook, or retrieve the run until its status is completed or failed:
async function wait(run_id: string) {
for (;;) {
const run = await a22a.runs.retrieve(run_id);
if (run.status === "completed" || run.status === "failed") return run;
await new Promise((resolve) => setTimeout(resolve, 2000));
}
}Automatic retries in the library
The library does not retry failed requests. Retrying requests lists which ones to retry yourself.
Webhook endpoint and balance in the API
The webhook endpoint and its signing secret are set on Webhook in the A22A dashboard, and your balance is shown on Plan. Neither is available with an API key yet.
Longer answers
A step's answer is capped at 4,096 tokens, or the model's maximum if lower. There is no setting to raise it yet. Split long work across steps.
Deleting data
Runs and projects can only be deleted along with your account. See Data and storage.
Checking without deploying
There is no way to check workflows without deploying them. Run a22a deploy only from your main branch; see Deploying from CI.
Environments
Every API key on an account deploys to the same projects. There is no separate staging. To try a change, deploy it as a project with another name, such as acme-staging.
OpenAPI spec
There is no OpenAPI document for the API yet. The API reference lists every endpoint, request and response.