openclaw - 💡(How to fix) Fix [Bug]: [REGRESSION] [Fix] 2026.5.12 SSRF protection blocks private/CGNAT model provider endpoints by default, breaking self-hosted LMStudio/vLLM/Ollama setups

Log Evidence
SSRF Block (repeating every agent run)

warn security {"subsystem":"security"} blocked URL fetch (url-fetch) targetOrigin=http://100.77.77.111:8877/ reason=Blocked hostname or private/internal/special-use IP address
warn provider-transport-fetch {"subsystem":"provider-transport-fetch"} [model-fetch] error provider=lmstudio api=openai-completions model=qwen/qwen3.6-35b-a3b elapsedMs=2 name=SsrFBlockedError code=undefined causeName=undefined causeCode=undefined message=Blocked hostname or private/internal/special-use IP address
text

### Cascading Agent Failure
 
warn agent/embedded {"subsystem":"agent/embedded"} {"event":"embedded_run_agent_end","tags":["error_handling","lifecycle","agent_end","assistant_error"],"isError":true,"error":"LLM request failed: network connection error.","failoverReason":"timeout","providerRuntimeFailureKind":"timeout"}
warn model-fallback/decision {"subsystem":"model-fallback/decision"} {"decision":"candidate_failed","fallbackStepFinalOutcome":"chain_exhausted","reason":"timeout"}
error diagnostic {"subsystem":"diagnostic"} lane task error: error="FailoverError: LLM request failed: network connection error."
error Embedded agent failed before reply: LLM request failed: network connection error.
text
 
  
 
 

### Cron Job Failure + Backoff

warn cron {"module":"cron"} {"jobId":"67aecddf-...","jobName":"HOURLY LOGS","error":"FailoverError: LLM request failed: network connection error."} cron: job run returned error status
info cron {"module":"cron"} {"consecutiveErrors":3,"backoffMs":300000} cron: applying error backoff
text
 
## Broken Config Keys Attempted

The 2026.5.12 config schema rejects every intuitive key a user might try to bypass this. Here is the full list of attempts and their validation errors:

| Attempted Key | Error |
|---|---|
| `<root>.dangerousAllowPrivateUrls` | `Config validation failed: <root>: Unrecognized key: "dangerousAllowPrivateUrls"` |
| `gateway.dangerousAllowPrivateUrls` | `Config validation failed: gateway: Unrecognized key: "dangerousAllowPrivateUrls"` |
| `<root>.security` | `Config validation failed: <root>: Unrecognized key: "security"` |
| `gateway.security` | `Config validation failed: gateway: Unrecognized key: "security"` |
| `<root>.fetch` | `Config validation failed: <root>: Unrecognized key: "fetch"` |
| `gateway.fetch` | `Config validation failed: gateway: Unrecognized key: "fetch"` |
| `gateway.ssrf` | `Config validation failed: gateway: Unrecognized key: "ssrf"` |
| `gateway.providers` | `Config validation failed: gateway: Unrecognized key: "providers"` |
| `<root>.providers` | `Config validation failed: <root>: Unrecognized key: "providers"` |

None of these are documented anywhere. The only way to find the correct key is to dump the schema (`openclaw config schema`) and grep for SSRF-related fields.

## Working Fix

The correct per-provider config key, found only by dumping the schema:


openclaw config set models.providers.lmstudio.request.allowPrivateNetwork true
openclaw gateway restart

The schema description for this field reads:

    "When true, allow HTTPS to the model base URL when DNS resolves to private, CGNAT, or similar ranges, via the provider HTTP fetch guard (fetchWithSsrFGuard). OpenAI Responses WebSocket reuses request for headers/TLS but does not use that fetch SSRF path. Use only for operator-controlled self-hosted OpenAI-compatible endpoints (LAN, overlay, split DNS). Default is false."

Verification

After applying the fix, logs show:

     ✅ Zero SsrFBlockedError messages
     ✅ Zero FailoverError messages
     ✅ Successful model completions: info channels/slack delivered reply to channel:C0B254QFYNN
     ✅ Gateway hot-reload picked up the change: info gateway/reload config hot reload applied (models.providers.lmstudio.request)

---

openclaw config set models.providers.lmstudio.request.allowPrivateNetwork true
openclaw gateway restart