fix stream corruption and timeout

add heartbeat to prevent timeout

fix blank page and keep alive

fix streaming and sync

streaming+sync

num_predict=-1 (unlimited), keep model loaded forever

fix: persistent chat history, reduced max_tokens, extended keep_alive

fix: export ollama vars + use Python proxy for chat

fix: nginx symlink + simplify Ollama API body

fix: add model name directly in JavaScript API call

fix: use deepseek-r1:8b model name in chat requests

fix: add zstd dependency for ollama install

minimal base image (python:3.11-slim-bookworm) + deepseek-r1:8b

switch to deepseek-r1:8b for 128K stability

fix: runtime nginx config with 600s timeout + direct Ollama proxy

fix: 504 timeout (300s nginx) + BrokenPipe handling

feat: remove OpenClaw gateway – use Ollama directly for chat

fix: chat via Ollama native API directly, bypass gateway

fix: call Ollama directly on port 11434, bypass gateway for chat

fix: proxy chat requests through health server, avoid CORS+gateway OPTIONS issue

fix: clean health_server.py with proper CORS/OPTIONS

fix: CORS/OPTIONS support for chat API calls

fix: clean start.sh with no sed commands

fix: clean Dockerfile with /v1/ proxy, OpenAI-compatible API

fix: Ollama direct API + nginx proxy

fix: route /v1/ to gateway for chat API

fix: correct API endpoint path

fix: stable mode:none + custom chat UI replacing broken canvas

fix: nginx injects Authorization header for canvas auth

fix: token auth + dangerouslyDisableDeviceAuth

fix: auth mode none – open canvas

fix: IQ3_XS model, clean Dockerfile, dynamic token

fix: revert Dockerfile to clean version

fix: IQ3_XS 14B (6.38GB) for 128K stability, auto-token auth

fix: nginx injects Bearer token for canvas API requests

fix: add --token to gateway + controlUi insecure auth for canvas

fix: add auth mode + remote token + env var for canvas access

fix: proper auth object with embedded token

fix: disable gateway auth explicitly

fix: clean start.sh - use health_server.py file only, no inline HTML

fix: standalone health server script avoids shell quoting issues

fix: remove auth, use openai-completions API for no-tool text mode, 128K context

fix: remove invalid thinking key, correct env var, 128K in health page

fix: canvas auth via token param, 128K context, thinking=high

fix: text-only model config to prevent tool errors, nginx routing for canvas

fix: official ollama model with tool support, health page on :8080, 96K context

fix: disable OpenClaw health-monitor via channelHealthCheckMinutes=0

web-only: remove Telegram, fixed auth token, kill health-monitor

fix: ollama provider api key, remove unrecognized identity block

fix: openclaw gateway mode=local, allow-unconfigured