Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

The voice-call plugin's streaming/conversation embedded response path ignores a routed voice agent's tools.allow: [] / no-tool configuration, sends tool schemas to an Ollama response model anyway, and times out before producing a spoken response.

Steps to reproduce

1. Run OpenClaw 2026.5.7 with @openclaw/voice-call 2026.5.7 configured for Twilio inbound calls, streaming transcription enabled, and conversation mode.
2. Configure a dedicated voice agent with a fast Ollama model and no tools, for example:

   {
     "id": "voice",
     "model": { "primary": "ollama/qwen2.5:1.5b", "fallbacks": [] },
     "thinkingDefault": "off",
     "reasoningDefault": "off",
     "fastModeDefault": true,
     "params": { "temperature": 0.2, "maxTokens": 80, "cacheRetention": "none" },
     "tools": { "allow": [] },
     "systemPromptOverride": "You are a fast phone voice assistant. Reply only as valid JSON: {\"spoken\":\"...\"}. Keep spoken under 18 words. No tool use."
   }

3. Route the voice-call plugin/number to that agent and model:

   {
     "plugins": {
       "entries": {
         "voice-call": {
           "config": {
             "responseModel": "ollama/qwen2.5:1.5b",
             "responseTimeoutMs": 20000,
             "numbers": {
               "+REDACTED": {
                 "agentId": "voice",
                 "responseModel": "ollama/qwen2.5:1.5b",
                 "responseTimeoutMs": 20000
               }
             }
           }
         }
       }
     }
   }

4. Call the Twilio number from an allowed caller and speak a short utterance, e.g. Hey, how you doing?.
5. Observe that STT succeeds and handleInboundResponse() starts, but no AI response is spoken; the embedded run times out and the call is ended after stream disconnect grace.
6. Compare with a direct Ollama /api/chat request using the same model and no tools; the same simple prompt returns in ~2 seconds.

Expected behavior

When the routed voice agent has tools.allow: [] and the prompt says no tool use, the voice-call embedded response run should be LLM-only or otherwise respect the agent's no-tool configuration. The request to Ollama should not include OpenClaw tool schemas, and the model should return a short spoken JSON response within the configured timeout.

Actual behavior

The voice-call embedded run compiles 13 OpenClaw tools despite the routed voice agent having tools.allow: []. The compiled trajectory shows tools such as image, image_generate, memory_get, memory_search, session_status, sessions_history, sessions_list, sessions_send, sessions_spawn, sessions_yield, subagents, update_plan, and video_generate being included. The Ollama request times out at responseTimeoutMs=20000, producing no assistant text, and voice-call logs Response generation error: Response generation was aborted.

OpenClaw version

2026.5.7 (eeef486)

Operating system

Linux 6.8.0-111-generic (x64)

Install method

npm global / OpenClaw Gateway running as systemd user service

Model

ollama/qwen2.5:1.5b for the voice response model

Provider / routing chain

Twilio Programmable Voice -> Tailscale Funnel -> OpenClaw voice-call webhook -> OpenAI streaming transcription -> OpenClaw embedded voice agent response -> local Ollama qwen2.5:1.5b -> OpenAI TTS

Additional provider/model setup details

• Voice-call plugin: @openclaw/voice-call 2026.5.7
• Voice-call mode: Twilio inbound, streaming transcription enabled, realtime disabled, conversation mode
• STT provider: OpenAI gpt-4o-transcribe
• TTS provider: OpenAI gpt-4o-mini-tts
• Dedicated routed voice agent: agentId: voice
• Voice agent model: ollama/qwen2.5:1.5b
• Voice agent config includes tools.allow: []
• Voice-call number route also sets agentId: voice and responseModel: ollama/qwen2.5:1.5b
• Direct Ollama /api/chat to qwen2.5:1.5b with the same short prompt works in ~2.2s when no tools are supplied.
• Direct Ollama /api/chat with the same 13 OpenClaw tool schemas takes materially longer (~7.8s in a warmed direct test) and the observed OpenClaw embedded run timed out at 20s during the real voice path.

Logs, screenshots, and evidence

# Gateway log excerpt, redacted
[voice-call] Transcript for <twilio-call-sid>: Hey, how you doing? (chars=19)
[voice-call] Auto-responding to inbound call <call-id>: "Hey, how you doing?"
[voice-call] Media stream disconnected: <twilio-call-sid> (<stream-sid>)
[voice-call] Auto-ending call <call-id> after stream disconnect grace
2026-05-07T23:24:58.096-07:00 [agent/embedded] embedded run timeout: runId=voice:<call-id>:1778221476764 sessionId=4f36d284-b20b-4f27-a7eb-b4209c0282a1 timeoutMs=20000
2026-05-07T23:24:58.115-07:00 [agent/embedded] embedded run failover decision: runId=voice:<call-id>:1778221476764 stage=assistant decision=surface_error reason=timeout from=ollama/qwen2.5:1.5b profile=-
[voice-call] Response generation error: Response generation was aborted

# Voice trajectory evidence from ~/.openclaw/agents/voice/sessions/<session>.trajectory.jsonl
session.started:
  agentId: voice
  messageProvider: voice
  provider: ollama
  modelId: qwen2.5:1.5b
  toolCount: 13

context.compiled:
  systemPrompt: You are a fast phone voice assistant. Reply only as valid JSON: {"spoken":"..."}. Keep spoken under 18 words. No markdown. No tool use. Be direct, warm, and conversational.
  prompt: Hey, how you doing?
  messages: []
  tools included despite tools.allow=[]:
    image, image_generate, memory_get, memory_search, session_status,
    sessions_history, sessions_list, sessions_send, sessions_spawn,
    sessions_yield, subagents, update_plan, video_generate

model.completed:
  aborted: true
  timedOut: true
  promptError: request timed out | request timed out
  assistantTexts: []
  finalPromptText: Hey, how you doing?

# Direct Ollama control test, same model, no tools
POST http://127.0.0.1:11434/api/chat model=qwen2.5:1.5b
elapsed: ~2.2s
response content: "I'm great, thanks for asking! How can I help today?"

# Direct Ollama comparison with the same 13 tool schemas included
payload chars: ~16082, tools: 13
elapsed: ~7.8s
response returned successfully in direct test, but the real embedded voice run timed out at 20s.

Impact and severity

Affected: Users running voice-call streaming/conversation mode with a routed low-latency local/Ollama voice agent configured with no tools.

Severity: High for that setup. STT succeeds, but the voice assistant produces no spoken response after the caller speaks.

Frequency: Observed on the tested inbound call path with the routed voice agent and ollama/qwen2.5:1.5b response model.

Consequence: Calls appear connected and transcription works, but the assistant goes silent after user speech; calls can then be ended by disconnect handling, making the voice responder unusable for this configuration.

Additional information

The local source appears to show why this happens. In the installed @openclaw/voice-call 2026.5.7 response-generator bundle, generateVoiceResponse() calls agentRuntime.runEmbeddedPiAgent({ ... }) with agentId, provider, model, timeoutMs, etc., but does not appear to pass an explicit toolsAllow from the routed agent config or disableTools: true for no-tool voice runs.

This looks related to the existing voice-call embedded-run plumbing issues, but is not the same symptom:

• #60118 — voice-call responseModel config ignored / model override plumbing
• #56367 — voice-call ignores agent sandbox defaults
• #71262 — realtime voice tool-bridging gap
• #17613 — embedded local LLM timeout while direct local request works

Possible fix direction: when voiceConfig.agentId routes to an agent with tools.allow, carry that allowlist into runEmbeddedPiAgent() as toolsAllow. If the effective allowlist is an empty array, pass disableTools: true or otherwise ensure no tool schemas are compiled into the embedded voice response run.