OpenClaw worked fine in 5.7 but is sluggish to receive CLI input, does not reply, and more

[turbid dove]

Does not honor model-changes, shoots back "assistant turn failed before producing content", HTTP 500: Internal Server Errors, gateway slow to restart, tui slow to start, Error logs are: 6:27:26 PM
warn
agent/embedded
low context window: openrouter/deepseek/deepseek-v4-flash ctx=200000 (warn<209715) source=agentContextTokens
6:27:27 PM
info
gateway/ws
⇄ res ✓ logs.tail 483ms conn=82d8db5b…03ac id=f5babdc4…657e
6:27:27 PM
warn
agent/embedded
workspace bootstrap file AGENTS.md is 14304 chars (limit 12000); truncating in injected context (sessionKey=agent:main:main)
6:27:27 PM
warn
agent/embedded
workspace bootstrap file TOOLS.md is 17424 chars (limit 12000); truncating in injected context (sessionKey=agent:main:main)
6:27:27 PM
warn
agent/embedded
workspace bootstrap file MEMORY.md is 15302 chars (limit 12000); truncating in injected context (sessionKey=agent:main:main)
6:27:31 PM
warn
agent/embedded
embedded run agent end
6:27:35 PM
warn
agent/embedded
embedded run agent end
6:27:42 PM
warn
agent/embedded
embedded run agent end
6:27:52 PM
warn
agent/embedded
embedded run agent end
6:27:52 PM
warn
agent/embedded
embedded run failover decision
6:27:52 PM
error
diagnostic
lane task error: lane=main durationMs=26303 error="FailoverError: HTTP 500: Internal Server Error"
6:27:52 PM
error
diagnostic
lane task error: lane=session:agent:main:main durationMs=26308 error="FailoverError: HTTP 500: Internal Server Error"
6:27:52 PM
warn
model-fallback/decision
model fallback decision
6:27:53 PM
error
Transient HTTP provider error before reply (HTTP 500: Internal Server Error). Retrying once in 2500ms.
Transient HTTP provider error before reply (HTTP 500: Internal Server Error). Retrying once in 2500ms.
6:27:56 PM
warn
agent/embedded
low context window: openrouter/deepseek/deepseek-v4-flash ctx=200000 (warn<209715) source=agentContextTokens

...this didn't happen with 5.7

[sage blade]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[onyx blaze]

First thing: can you paste openclaw --version and openclaw status --all? Version mismatch/regression is the fastest way to explain “worked on 5.7, broken now.”

From those logs, I see two separate things:

1. The actual reply failure is upstream/provider-shaped: openrouter/deepseek/deepseek-v4-flash is returning HTTP 500: Internal Server Error, then OpenClaw is entering model fallback. That can produce “assistant turn failed before producing content” if every attempted candidate fails before a reply starts.

2. The bootstrap warnings are probably contributing to sluggishness/prompt pressure, but they’re not themselves fatal. AGENTS.md, TOOLS.md, and MEMORY.md are all over the default 12k per-file injected-context limit, so every run is carrying/truncating a lot of workspace instructions. Current docs say this is controlled by:

{
  agents: {
    defaults: {
      contextInjection: "continuation-skip",
      bootstrapMaxChars: 12000,
      bootstrapTotalMaxChars: 60000
    }
  }

[onyx blaze]

}

Docs: https://docs.openclaw.ai/gateway/configuration-reference#agentsdefaultscontextinjection

For immediate triage, I’d run:

openclaw --version
openclaw status --all
openclaw models status --probe
openclaw doctor

If models status --probe shows OpenRouter/DeepSeek failing but another configured provider is healthy, switch the default/fallback with:

openclaw models set <provider/model>

[onyx blaze]

Docs: https://docs.openclaw.ai/cli/models

If model changes still don’t stick after that, that’s likely worth a GitHub issue with the version, status --all, and these log lines, because session model switching/fallback coordination is explicitly a runtime behavior now: https://docs.openclaw.ai/concepts/model-failover#session-overrides-and-live-model-switching

[turbid dove]

I need to downgrade back to 5.7 ..how?

[onyx blaze]

Use the updater with an exact tag:

openclaw --version
openclaw update status
openclaw update --tag 2026.5.7 --dry-run
openclaw update --tag 2026.5.7

It should prompt because this is a downgrade. If you’re sure and want non-interactive:

openclaw update --tag 2026.5.7 --yes

If your install is a direct global package install and the updater path fails, the package version exists on npm, so the fallback is:

[onyx blaze]

pnpm add -g openclaw@2026.5.7
openclaw gateway restart

Docs: https://docs.openclaw.ai/cli/update and https://docs.openclaw.ai/install/development-channels

[peak chasm]

openclaw gateway restart
Config was last written by a newer OpenClaw (2026.5.12); current version is 2026.5.7.

🦞 OpenClaw 2026.5.7 (eeef486)
   I'm not magic—I'm just extremely persistent with retries and coping strategies.

Gateway restart blocked: Refusing to restart the gateway service because this OpenClaw binary (2026.5.7) is older than the config last written by OpenClaw 2026.5.12.
Tip: Run the newer openclaw binary on PATH, or reinstall the intended gateway service from the newer install.
Tip: Set OPENCLAW_ALLOW_OLDER_BINARY_DESTRUCTIVE_ACTIONS=1 only for an intentional downgrade or recovery action.

[onyx blaze]

That block is expected for an intentional downgrade. The 5.7 binary sees config touched by 5.12 and refuses service actions unless you explicitly allow the recovery action.

For one command:

OPENCLAW_ALLOW_OLDER_BINARY_DESTRUCTIVE_ACTIONS=1 openclaw gateway restart

If that still refuses because the service itself points at the wrong install, reinstall the service from the downgraded binary, then restart:

OPENCLAW_ALLOW_OLDER_BINARY_DESTRUCTIVE_ACTIONS=1 openclaw gateway install --force
OPENCLAW_ALLOW_OLDER_BINARY_DESTRUCTIVE_ACTIONS=1 openclaw gateway restart

Leave that env var unset afterward. It’s specifically for intentional downgrade/recovery, not normal use.