#Using Anthropic MAx Plan with Open CLaw

1 messages · Page 1 of 1 (latest)

tulip flint
#

anyone ?

glacial flameBOT
#

Claude Max (or Pro) doesn’t come with an Anthropic API key, so trying to “reuse the web app” via unofficial proxies (like claude-serve) is exactly the kind of thing that can get accounts flagged/banned — and we can’t help with ToS-evading workarounds.

The supported way in OpenClaw for subscription accounts is Anthropic setup-token:

  1. Install/use Claude Code CLI (on any machine) and generate a token:
claude setup-token
  1. Paste it into OpenClaw (either via onboarding wizard, or CLI):
openclaw models auth paste-token --provider anthropic
# or generate it on the gateway host:
openclaw models auth setup-token --provider anthropic
  1. Verify:
# 
openclaw models status

Docs (these sections answer your exact question):

If you paste:

  • openclaw --version
  • the exact “not authorised” error text
  • openclaw models status (redacts secrets automatically)

…I can help you pinpoint why it’s failing (wrong token pasted, provider not set as default, gateway using a different config, etc.).

tulip flint
#

anthropic effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=1, api_key=0) | anthropic:default=token:sk-ant-o it is already settup as you sais but every time I try to use Claude Opus as the model, it goes to the fallback models.

#

?

tulip flint
#

@glacial flame ?

glacial flameBOT
#

Yep, I’m here.

If you’re hitting “not authorised” with Claude Max, the key thing is: don’t use claude-serve/unofficial proxies. Use the Anthropic setup-token flow instead (that’s the supported “subscription auth” path in OpenClaw).

To debug your specific case, paste all three (safe to share):

  1. openclaw --version
  2. openclaw models status
  3. The exact error line(s) you see in logs (or run openclaw status --all and paste the relevant chunk)

Then I can tell you exactly what’s misconfigured.

tulip flint
#

no i am using the direct auth token with open claw but it doesnt work @glacial flame

glacial flameBOT
#

If you’re already using the Anthropic setup-token path and it still “doesn’t work”, the next step is to pin down which “not authorised” you mean — OpenClaw can say that for gateway auth or for Anthropic auth, and the fixes are totally different.

1) First: confirm the token type

When you say “direct auth token”, it needs to be the one generated by Claude Code:

claude setup-token

Not claude login, not a web cookie, not a proxy token.

2) Make sure you pasted it on the gateway host

Auth profiles live on the machine running the Gateway. If your Gateway is on a VPS but you pasted the token on your laptop, the Gateway won’t see it.

3) Run these and paste outputs (they redact secrets)

# 
openclaw --version
openclaw models status --probe --probe-provider anthropic
openclaw status --all

4) Also paste the exact error text

Specifically, is it:

  • Control UI says “unauthorized” / keeps reconnecting → that’s gateway.auth.token / dashboard auth, not Anthropic.
  • Logs show Anthropic 401 / “not_authorized” → token is invalid/expired/revoked, or the wrong profile is being used.

Once you paste those outputs + the exact error line, I can tell you which one it is and what to change. Docs reference: https://docs.openclaw.ai/help/faq#do-you-support-claude-subscription-auth-claude-pro-or-max

tulip flint
#

While I figure that out, can you tell me the options for the best real-time voice agent set up for OpenClaw? I want to be able to talk to OpenClaw in real time. @glacial flame

glacial flameBOT
#

If “talk to OpenClaw in real time” is the goal, there are basically three tiers you can pick from:

1) Best “real-time” experience: Talk Mode (macOS / iOS / Android)

This is the continuous loop (listen → transcript → model → speak) with interrupt-on-speech etc. It’s the closest thing to a live voice assistant UX in OpenClaw right now.

2) Actual phone calls: voice-call plugin (Twilio/Telnyx/Plivo)

If you want to call your agent (or have it call you) and talk over the phone network:

(There’s also a community Vapi ↔ OpenClaw bridge showcased here: https://docs.openclaw.ai/start/showcase#-voice--phone)

#

3) Easiest setup (not streaming, but practical): voice notes + transcription + TTS replies

Works well in Discord/Telegram/etc: you send a voice message, OpenClaw transcribes it, and it replies with a voice note.

If you tell me what device you’re on (macOS app? iPhone? Android? Linux server only?) and whether you mean live conversation vs phone calls, I can point you at the best path with the least pain.

tulip flint
#

@glacial flame what is the best alternative to memory lance db plugin as it seems to use openAI API and i worry about the costs

glacial flameBOT
#

You don’t have to use the memory-lancedb plugin (and you don’t need OpenAI) just to get useful memory.

Best “no-surprise-cost” alternative

Use the default memory-core (Markdown files + local index) and run embeddings locally:

  • Switch memory slot away from LanceDB:
{
  plugins: {
    slots: { memory: "memory-core" } // or "none" to disable memory plugins entirely
  }
}
  • Then make memory search local (no API spend):
# 
{
  agents: {
    defaults: {
      memorySearch: {
        provider: "local",
        fallback: "none"
      }
    }
  }
}

Check what it’s doing with:

openclaw memory status --deep