Summary

Currently, OpenClaw's kimi provider cannot send images to Kimi Code Plan API (api.kimi.com/coding/v1). When a user attaches an image in the conversation, the request will fail.

Problem to solve

The request fails with: HTTP 403 Forbidden {"error":{"message":"Kimi For Coding is currently only available for Coding Agents such as Kimi CLI, Claude Code, Roo Code, Kilo Code, etc.","type":"access_terminated_error"}} This happens because:

1. OpenClaw uses User-Agent openclaw/* which is not in Kimi's whitelist
2. The current endpoint configuration may not be optimal for image requests

Proposed solution

Update the kimi provider to use Kimi-approved User-Agent and endpoint: Current behavior (broken for images):

• Endpoint: /messages (or inferred from api: "openai-completions")
• User-Agent: openclaw/2026.4.27
• Result: 403 Forbidden when sending images Proposed changes:

1. Use /chat/completions endpoint for OpenAI-format requests
   • Kimi Code Plan supports OpenAI Chat Completions format at /chat/completions
   • Keeps existing openai-completions API format unchanged
   • No need to switch to Anthropic Messages format
2. Set User-Agent to a Kimi-whitelisted value
   • Kimi strictly validates User-Agent against an allowlist
   • Tested working values:
     • claude-code/0.1.0 ✅
     • claude-code/1.0.0 ✅
     • Kilo-Code/1.0 ✅
   • Tested failing values:
     • openclaw/* ❌
     • kimi-cli/1.0 ❌ (surprisingly!)
     • Roo-Code/1.0 ❌
   • Recommended: claude-code/0.1.0 (most compatible, tested extensively)
3. Keep standard OpenAI image format

   {
     "type": "image_url",
     "image_url": {
       "url": "data:image/jpeg;base64,..."
     }
   }

Alternatives considered

Impact

Affected users/systems/channels:

• All OpenClaw users who have configured the kimi provider with a Kimi Code Plan API key (sk-kimi-*)
• Users who attach images (screenshots, diagrams, photos) in OpenClaw conversations expecting the AI to analyze them
• Specifically impacts the kimi-code model in the kimi provider configuration

Severity:

• Blocks workflow for image-related tasks. Users cannot use Kimi Code Plan for any vision-based tasks within OpenClaw, including:
  • Analyzing error screenshots
  • Reviewing UI/UX designs
  • Reading diagrams or charts
  • Processing photographed documents

Frequency:

• Always — Every time a user attaches an image while using the kimi-code model, the request fails with 403 Forbidden. This is 100% reproducible.

Consequence:

• Users must switch to other providers (e.g., OpenAI, Anthropic) for image analysis, defeating the purpose of having Kimi Code Plan configured
• Forces users to maintain multiple API keys and providers just to handle images
• Creates confusion because the kimi-code model is advertised as supporting ["text", "image"] input, but images silently fail
• Extra manual work: users must download images and upload them to Kimi's web interface or other tools instead of using OpenClaw's integrated chat

Evidence/examples

This fails (current OpenClaw behavior) curl https://api.kimi.com/coding/v1/chat/completions
-H "Authorization: Bearer $KIMI_API_KEY"
-H "Content-Type: application/json"
-H "User-Agent: openclaw/2026.4.27"
-d '{ "model": "kimi-code", "messages": [{"role": "user", "content": [{"type": "image_url", "image_url": {"url": "data:image/jpeg;base64,..."}}, {"type": "text", "text": "Describe this"}]}] }' → 403 Forbidden This works (proposed fix) curl https://api.kimi.com/coding/v1/chat/completions
-H "Authorization: Bearer $KIMI_API_KEY"
-H "Content-Type: application/json"
-H "User-Agent: claude-code/0.1.0"
-d '{ "model": "kimi-code", "messages": [{"role": "user", "content": [{"type": "image_url", "image_url": {"url": "data:image/jpeg;base64,..."}}, {"type": "text", "text": "Describe this"}]}] }' → 200 OK with image description

Additional information

Additional Context

• API Base URL: https://api.kimi.com/coding/v1
• Required Headers: Authorization, Content-Type, User-Agent: claude-code/0.1.0
• No anthropic-version header needed when using /chat/completions
• Image format: Standard OpenAI image_url with base64 data URL
• Tested image size: ~1MB JPEG works correctly
• Model ID: kimi-code

Environment

• OpenClaw version: 2026.4.27
• OS: Ubuntu 22.04
• Kimi Code Plan API key format: sk-kimi-*