Repository navigation
docs: document UberLLM OpenAI-compatible gateway setup #3375
Description
Proposal
Add a short UberLLM section to docs/local-openai-compatible-providers.md, after the LLMTR one (#3373). Disclosure: I work on UberLLM.
UberLLM is a pay-as-you-go OpenAI-compatible gateway at https://api.uberllm.dev/v1 (prepaid, no free tier), with Bearer keys (ull_...). Claw already works with it through OPENAI_BASE_URL and OPENAI_API_KEY. The docs would cover two things:
- Model ids are vendor-prefixed (
deepseek/deepseek-v4-flash,moonshotai/kimi-k2.6), so they need thelocal/hatch:local/deepseek/deepseek-v4-flash. Otherwise a baredeepseek/-style name could be taken as a native route. - Claw asks for up to 64,000 output tokens. Each model's limit is the
max_output_tokensfield of the publicGET /v1/modelslist (no key needed), so you can pick a model that allows at least that much.
Checked with the v0.1.11 release binary on Linux: claw --model local/deepseek/deepseek-v4-flash prompt "Reply exactly HELLO_WORLD_123" printed HELLO_WORLD_123, and so did local/moonshotai/kimi-k2.6.
I'll open a docs-only PR that references this issue, unless you'd rather not list it.
🤖 Generated with Claude Code
Activity
PSA-Team-source commentedon Nov 4, 2026
Correction after more testing with v0.1.11: most bare UberLLM ids already work (
deepseek/deepseek-v4-flash,moonshotai/kimi-k2.6andopenai/gpt-5-miniall printedHELLO_WORLD_123). The one that breaks isqwen/qwen3.8-flash, which goes to DashScope (missing DashScope credentials), whilelocal/qwen/qwen3.8-flashworks. So the docs will just say to uselocal/<catalog id>for every UberLLM id, which always sends the id unchanged.XinwangTao commentedon Nov 5, 2026
Thanks for the thorough proposal and for verifying against the v0.1.11 binary. The merged docs (#3376) make the
local/<catalog id>prefix guidance clear, especially the qwen/qwen3.8-flash case that falls back to DashScope without the prefix. Closing this out as completed.