OpenClaw

OpenClaw "Agent database execution admission is closed"

Last checked

The error

Agent database execution admission is closed

SQLite worker stopped before its result was received: Worker terminated due to reaching memory limit: JS heap out of memory | ERR_WORKER_OUT_OF_MEMORY

The first line is what the turn fails with. The second line is in the Gateway log just before it. In chat you may only see "OpenClaw couldn't finish this request" or "This Codex session changed before your message could run. Please send it again."

OpenClaw keeps each agent's session history in a SQLite database that is read by a worker thread with a hard-coded 512 MB memory cap. A very large session (the report on openclaw/openclaw#191847 had about 127,000 transcript events and a 1.06 GB database) pushes that worker past the cap. The worker is killed and recreated, the database stops admitting new work, and every turn in that session fails. Start a new session for the same agent, archive the oversized one, or roll back from a pre-update backup. As of 10 November 2026 no release fixes it, including 2026.11.2.

Why it happens

The SQLite worker that loads session history has a fixed memory cap (maxOldGenerationSizeMb defaults to 512 in createWorkerTaskHost, with no config override). Loading a very large session exceeds it, the worker dies with ERR_WORKER_OUT_OF_MEMORY, and OpenClaw closes database admission for that session until the worker is back. The next turn hits the same wall.

  1. One very large session. The reporter of issue #191847 upgraded a Raspberry Pi 5 from 2026.9.7 to 2026.10.10 with a session of about 127k events and a 1.06 GB agent database (WAL). The worker was recreated 13 times and 11 of 11 turns failed.
  2. The 512 MB worker cap is hard-coded. There is no setting to raise it, so a bigger host does not help on its own.
  3. Bounded session loading is not in a release yet. A maintainer comment on the issue notes that 2026.11.1 ships the bounded-hydration change from PR #172970 but not a fix for this crash, and a synthetic reproduction runs out of memory on both 2026.10.10 and 2026.11.1.

The fix

  1. 1 Confirm it is this bug: search the Gateway log for ERR_WORKER_OUT_OF_MEMORY just before each failed turn.
  2. 2 Start a new, separate session for the same agent. This is the workaround that worked for users on the issue. /new inside the broken session did not help, because the session id stayed the same.
  3. 3 Find the oversized session: check the size of the agent database in your OpenClaw state directory. A database near or above 1 GB is the warning sign.
  4. 4 Archive or trim old sessions you no longer need so the agent's active history stays small.
  5. 5 If you need the old session working today, roll back to the version you came from (for example 2026.9.7) using the backup taken before the update. Older releases cannot open newer database schemas, so a plain downgrade without that backup can make things worse.
  6. 6 Watch openclaw/openclaw#191847 and its companion #191273 for the release that ships the fix, then update.
openclaw gateway status

What did not help in the report

Switching the agent's harness from codex to claude-cli did not change anything, and neither did /new, because both kept the same session id.

A Gateway restart and a full host reboot also did not help. The worker hits the same cap as soon as it loads the session again.

One user raised the cap from 512 to 2048 MiB in the source as a temporary local patch. That is an edit to OpenClaw's code, not a setting, and it is lost on the next update.

When to roll back

Roll back only if the oversized session matters more than the fixes you would give up, and only from a backup taken before the update. The update guide covers the rollback steps.

Still failing?

  • Other agents fail too, not only one session. That is a different problem: check memory pressure on the host with the OOM guide.
  • A new session fails straight away. The database itself may be damaged. Restore the pre-update backup.
  • The turn fails with a Codex 403 instead. See the "cannot verify the owner" page below.

Related errors

Full guideOpenClaw Latest Version: 2026.11.2, What Changed, Is It SafeEvery error, one pageHow to update OpenClaw

Hit a different error?

Paste any agent error and get the cause and fix in seconds.

Open the decoder

Frequently asked questions

What does "Agent database execution admission is closed" mean in OpenClaw?

The worker thread that reads the agent's SQLite database ran out of memory and was restarted, so OpenClaw stopped admitting new database work for that session. Every turn in the session then fails until the session is smaller or replaced.

Does updating to OpenClaw 2026.11.2 fix it?

No. As of 10 November 2026, issue #191847 is still open and no release notes list a fix. 2026.11.1 added bounded history loading, but a reproduction still runs out of memory on it.

Will a bigger server or more RAM help?

Not on its own. The 512 MB cap is hard-coded on the worker thread, not taken from the machine's memory, so the worker dies at the same point on any host.

How do I keep using the agent right now?

Start a new, separate session for the same agent. That worked for users on the issue. Archive the old, oversized session rather than reopening it.

Can I roll back to 2026.9.7?

Yes, if you took a backup before updating. Newer releases change the database schema and older releases cannot open it, so restore the backup instead of only installing the older package.

Stop firefighting agent errors

Decoding errors one at a time is the manual version of what BetterClaw automates. Run your agents on a no-code AI agent platform with managed models, retries and config validation built in.

Deploy Your Agent — Start FreeBook a Demo

Free plan available · Pro $49/mo · BYOK · 7-day free trial