5 replies

voice-call plugin: EADDRINUSE when agent invokes `voice_call` tool

Our setup: We run the Gateway as a systemd system service (

User=XXX

User=XXX

) with secrets in a root-owned

EnvironmentFile

EnvironmentFile

(

/etc/clawdbot-gateway.env

/etc/clawdbot-gateway.env

, mode 600). This means env vars like

${TWILIO_FROM_NUMBER}

${TWILIO_FROM_NUMBER}

are available inside the service but not in interactive shell sessions. We're aware this is "against the grain" — wondering if this is contributing to the issue or if others see this too.

Specs: OpenClaw 2026.2.3 · Ubuntu 24.04 · Node v22.22.0 · voice-call plugin· Twilio

What's broken:

1. When our agent (Kimi, kimi-coding/k2p5) invokes the

voice_call

voice_call

tool, it fails with

EADDRINUSE

EADDRINUSE

on the webhook port (3335). The service already owns that port — same PID. The tool appears to be creating a second runtime instead of reusing the service's singleton.
2.

openclaw gateway call voicecall.initiate

openclaw gateway call voicecall.initiate

(RPC) worked once on the original gateway process, but after any

systemctl restart

systemctl restart

, it returns "unknown error". Built-in RPC methods (

status

status

health

health

) still work. Plugin shows

loaded

loaded

, webhook port is bound,

plugins doctor

plugins doctor

is clean.

What we've tried:
-

openclaw plugins doctor

openclaw plugins doctor

→ clean, no issues
-

openclaw plugins list

openclaw plugins list

→ voice-call appears once, bundled, v2026.2.3
- Tried

gateway call --url ws://127.0.0.1:18789 --token "$TOKEN"

gateway call --url ws://127.0.0.1:18789 --token "$TOKEN"

— always fails with "gateway url override requires explicit credentials" even with valid 44-char token
- Dropping

--url

--url

lets CLI auto-discover the gateway and the RPC worked (once, before restart)

Workaround (partial): Teaching agent to shell out to

openclaw gateway call voicecall.initiate --params '{"message":"...","mode":"notify"}' --json

openclaw gateway call voicecall.initiate --params '{"message":"...","mode":"notify"}' --json

instead of using the native

voice_call

voice_call

tool.

Questions:
1. Is our systemd EnvironmentFile setup known to cause issues with the voice-call tool/service singleton?
2. Are plugin-registered RPC methods (

voicecall.initiate

voicecall.initiate

) expected to survive a gateway restart?
3. Is

voice_call

voice_call

tool supposed to share the service's runtime, or create its own?

Ty!

FotC�

Friends of the Crustacean 🦞🤝•3d ago•

5 replies

voice-call plugin: EADDRINUSE when agent invokes `voice_call` tool

Our setup: We run the Gateway as a systemd system service (

User=XXX

User=XXX

) with secrets in a root-owned

EnvironmentFile

EnvironmentFile

(

/etc/clawdbot-gateway.env

/etc/clawdbot-gateway.env

, mode 600). This means env vars like

${TWILIO_FROM_NUMBER}

${TWILIO_FROM_NUMBER}

voice_call

voice_call

tool, it fails with

EADDRINUSE

EADDRINUSE

on the webhook port (3335). The service already owns that port — same PID. The tool appears to be creating a second runtime instead of reusing the service's singleton.
2.

openclaw gateway call voicecall.initiate

openclaw gateway call voicecall.initiate

(RPC) worked once on the original gateway process, but after any

systemctl restart

systemctl restart

, it returns "unknown error". Built-in RPC methods (

status

status

health

health

) still work. Plugin shows

loaded

loaded

, webhook port is bound,

plugins doctor

plugins doctor

is clean.

What we've tried:
-

openclaw plugins doctor

openclaw plugins doctor

→ clean, no issues
-

openclaw plugins list

openclaw plugins list

→ voice-call appears once, bundled, v2026.2.3
- Tried

gateway call --url ws://127.0.0.1:18789 --token "$TOKEN"

gateway call --url ws://127.0.0.1:18789 --token "$TOKEN"

— always fails with "gateway url override requires explicit credentials" even with valid 44-char token
- Dropping

--url

--url

lets CLI auto-discover the gateway and the RPC worked (once, before restart)

Workaround (partial): Teaching agent to shell out to

openclaw gateway call voicecall.initiate --params '{"message":"...","mode":"notify"}' --json

openclaw gateway call voicecall.initiate --params '{"message":"...","mode":"notify"}' --json

instead of using the native

voice_call

voice_call

tool.

Questions:
1. Is our systemd EnvironmentFile setup known to cause issues with the voice-call tool/service singleton?
2. Are plugin-registered RPC methods (

voicecall.initiate

voicecall.initiate

) expected to survive a gateway restart?
3. Is

voice_call

voice_call

tool supposed to share the service's runtime, or create its own?

Ty!

voice-call plugin: EADDRINUSE when agent invokes `voice_call` tool

Similar Threads

voice-call plugin: EADDRINUSE when agent invokes `voice_call` tool

Similar Threads

Similar Threads

Similar Threads