typebase-io-cli logs
Stream runtime logs from your deployed Typebase server.
Last updated on
Stream runtime logs from your deployed Typebase server. Press x or Ctrl+C to stop.
logs <target>
Stream the logs of your dev or prod deployment.
npx typebase-io-cli logs devnpx typebase-io-cli logs prodTo deploy and start streaming in one step, pass --logs to the deploy command instead:
npx typebase-io-cli deploy dev --logsEach line shows the time, the level, the request it belongs to, and the message:
12:34:56 INFO POST /rpc/mutations/todos/create 201
12:34:56 INFO added todo 42
12:34:57 ERROR Error: could not notify subscriberOptions
| Flag | Description | Default |
|---|---|---|
--provider <provider> | Deployment provider: vercel, cloudflare, or deno | From typebase.json |
The provider is resolved the same way as deploy: the --provider flag wins, then serverProvider in
typebase.json, and if neither is set the CLI asks you to pick one.
Stopping the stream
Press x, X, or Ctrl+C. The command restores your terminal on the way out, including when it is stopped mid-connection.
The keypress only works when the command owns a terminal. If you pipe the output somewhere (| tee, CI, a background job), x does nothing — send
SIGINT to the process instead.
What the providers have in common
- The stream is attached to the deployment currently serving that target's URL, so run it after a
deploy. If the target was never deployed, the command says so and exits with a non-zero code. - Connections are re-established automatically, so a stream can stay open indefinitely.
- Levels are normalized:
warnprints asWARNING, and anything Typebase treats as an error prints asERRORin red.
Provider differences
How logs reach you differs per provider, because each one exposes a different API:
| Behaviour | Vercel | Deno Deploy | Cloudflare |
|---|---|---|---|
| Delivery | Polled every 2s | Streamed (NDJSON) | Streamed (tail session) |
| Delay before a line | ~10–20s | ~1–2s | Under a second |
console.log calls | First one per request only | All | All |
| Requests with no log | Shown | Not shown | Shown |
| History on start | Last ~60s | Last ~60s | None, live only |
| Survives a redeploy | No, restart the command | No, restart the command | Yes |
Vercel
Vercel has no working log-streaming endpoint, so the CLI polls the same request-logs API the Vercel dashboard uses, every two seconds, and skips requests it has already printed.
Lines typically appear 10–20 seconds after the request. That is Vercel's log ingestion delay — the dashboard and the official vercel logs CLI are
equally delayed, so there is nothing the CLI can do to be faster.
If an action calls console.log several times while handling one request, Vercel's API only keeps the first one. Combine the values into a single
console.log when you need to see more than one.
Deno Deploy
Deno Deploy streams newline-delimited JSON, so lines arrive as they happen and every console.log is delivered. Lines produced while handling a
request carry a shortened OpenTelemetry trace id, which tells you which lines belong together:
12:34:56 INFO [a1b2c3d4] fetching feed
12:34:56 INFO [a1b2c3d4] found 12 postsThe Deno Deploy console splits Logs and Traces into separate tabs and only the Logs API is public. Requests that never call console.log therefore
do not appear in the stream. Paste a trace id into the console's Traces search to see the full request.
logs dev follows the specific preview revision that was live when it started. After deploying again, restart the command to follow the new one.
Cloudflare
Cloudflare logs arrive over a tail session — the same mechanism
wrangler tail uses, and the fastest of the three. One message covers a whole invocation: the request line, then everything it logged, then any
exception with its stack.
Because the tail follows the Worker by name, it keeps working across deploys: redeploy in another terminal and the stream picks up the new version.
Needs the global WebSocket, available from Node 22.4. On older versions the command exits with an error telling you to upgrade. Vercel and Deno
Deploy are unaffected.
Tail sessions expire after about an hour; the CLI opens a new one automatically. When you stop the stream it shows a short Closing the log stream... spinner while it releases the session on Cloudflare's side.
Cloudflare samples tail messages when a Worker takes more traffic than the session can carry, so a very busy Worker may not show every request. Dev and prod deployments with normal traffic stream everything.