error

[burnt garden]

on latest version of openclaw, the model shows followiong toek usage,
agent main | session main (openclaw-tui) | kimi-code/kimi-for-coding | tokens 116k/262k (44%)

however, i am getting this error:

HTTP 400: Invalid request: Your request exceeded model token limit: 262144 (requested: 347677)

how do i reset and get it to work again?

[dusk vapor]

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

[burnt garden]

@bronze sigil

[bronze sigil]

This happens when session context accumulates over time. Even though the visible count shows 44%, the actual request includes:

1. Visible tokens: 116k (your conversation)
2. System/system prompt tokens: Hidden context that adds up
3. Request overhead: The actual prompt being sent

Combined, this pushes you over the 262k limit → 347k requested.

Fix: Start a Fresh Session

The simplest solution: Start a new session/chat. This clears all accumulated context and resets the count.

In your TUI or client:

• Close the current session
• Start a fresh session with openclaw-tui

Prevent It Next Time

• Keep conversations focused on single tasks when possible
• Periodically start fresh sessions for new topics
• Be aware of very long conversations

Docs Reference

• Models docs: https://docs.moltbot.ai/concepts/models
• Kimi/Moonshot context limits are hard limits — once you hit them, you need to reset

---

Quick question: What openclaw --version are you running? Sometimes version mismatches can cause unexpected behavior with token counting. 🦐

[burnt garden]

openclaw --version
2026.1.29

is it latest version @bronze sigil

and the session that got rate limited was initated by telegram, how exactly to clear the session?

[bronze sigil]

Thanks! Version 2026.1.29 is current — that's not the issue.

The token counting discrepancy you saw is expected behavior with Moonshot/Kimi. Their limit is a hard cap, and once you hit it, you must start a fresh session.

Summary

| What you saw    | Reality                                 |
| --------------- | --------------------------------------- |
| 116k/262k (44%) | Only counts visible conversation tokens |
| HTTP 400        | Request exceeded 262k limit (was 347k)  |

Action Steps

1. Start a fresh session in openclaw-tui
2. Continue your work from there
   That's it. No config changes needed — just a new session clears the accumulated context.

Glad to help! 🦐

[bronze sigil]

Version is current — confirmed ✅

For Telegram sessions, the easiest way to get a fresh session:

1. Start a new conversation in Telegram — send a message to your bot. This typically creates a fresh session context.

2. If it's a persistent session, use the CLI:

   openclaw session list
   

   Find the Telegram session and clear it, or just restart openclaw entirely.

3. Quickest fix: Restart the openclaw process, then message from Telegram again.

---

The key point: fresh Telegram message = fresh session (unless you've got persistent sessions configured differently). Just try talking to the bot again in a new conversation thread. 🦐

[burnt garden]

HI please check openclaw crashes as soon i send a voice message from telegram, i can see in session log, that deepgram did transcribe it, but i think kimi 2.5 is sending alot of garbage text back, and causing openclaw to crash

@bronze sigil

[bronze sigil]

Oof, that's not good. A few questions to help diagnose:

Needed Info

1. openclaw --version (confirm you're still on 2026.1.29)

2. Crash logs — Run:

   clawdbot logs --tail=100
   

   Look for errors around the crash time.

3. Session logs — For the Telegram session:

   openclaw session logs