AIlice clearly values Chinese models — DeepSeek is a first-class deepseek: prefix, and the README calls GLM-4.5 the best open-source model, with Qwen 2.5 72B recommended for sub-agents. But right now the practical paths to those models are either OpenRouter pass-through (platform markup on top of provider rates) or direct signup with each provider (Chinese phone number + Alipay — a blocker for international users).
I built AnyLLM, an OpenAI-compatible API gateway with 24+ Chinese and international models behind a single key: DeepSeek V4, Qwen3.7-Plus, GLM-5.2, Kimi K2.5, MiniMax M3 — alongside GPT-5 and Gemini. International cards and crypto accepted, pass-through pricing with zero markup, built-in routing and failover.
Why this fits AIlice
Made for agentModelConfig. AIlice's core pattern is per-agent model assignment. With one AnyLLM endpoint, swapping a sub-agent from Qwen to GLM to Kimi is just a modelID change — no new signup, no new config block, no aggregator markup.
Works via the existing config path. config.json already accepts custom OpenAI-compatible endpoints (baseURL + apikey + contextWindow, the LM Studio pattern). AnyLLM drops in today with zero code changes.
IACT trees are token multipliers. Recursive task decomposition, bidirectional error correction, and multi-agent retries mean many LLM calls per task. Pass-through pricing on the Chinese tier decides how deep the decomposition can afford to go.
Failover for long autonomous runs. AIlice's fault tolerance lives at the agent level; AnyLLM adds provider-level failover beneath it — an IACT tree doesn't stall because one provider hiccuped mid-task.
Proposal
A short docs/README section: "Using an OpenAI-compatible gateway (AnyLLM)" with a config.json example following the existing custom-endpoint pattern. Happy to open a PR.
(Context: I'm the developer of AnyLLM — happy to provide a sandbox key for testing.)
Best,
Leo Bennett
leo.indiedev@gmail.com
AIlice clearly values Chinese models — DeepSeek is a first-class deepseek: prefix, and the README calls GLM-4.5 the best open-source model, with Qwen 2.5 72B recommended for sub-agents. But right now the practical paths to those models are either OpenRouter pass-through (platform markup on top of provider rates) or direct signup with each provider (Chinese phone number + Alipay — a blocker for international users).
I built AnyLLM, an OpenAI-compatible API gateway with 24+ Chinese and international models behind a single key: DeepSeek V4, Qwen3.7-Plus, GLM-5.2, Kimi K2.5, MiniMax M3 — alongside GPT-5 and Gemini. International cards and crypto accepted, pass-through pricing with zero markup, built-in routing and failover.
Why this fits AIlice
Made for agentModelConfig. AIlice's core pattern is per-agent model assignment. With one AnyLLM endpoint, swapping a sub-agent from Qwen to GLM to Kimi is just a modelID change — no new signup, no new config block, no aggregator markup.
Works via the existing config path. config.json already accepts custom OpenAI-compatible endpoints (baseURL + apikey + contextWindow, the LM Studio pattern). AnyLLM drops in today with zero code changes.
IACT trees are token multipliers. Recursive task decomposition, bidirectional error correction, and multi-agent retries mean many LLM calls per task. Pass-through pricing on the Chinese tier decides how deep the decomposition can afford to go.
Failover for long autonomous runs. AIlice's fault tolerance lives at the agent level; AnyLLM adds provider-level failover beneath it — an IACT tree doesn't stall because one provider hiccuped mid-task.
Proposal
A short docs/README section: "Using an OpenAI-compatible gateway (AnyLLM)" with a config.json example following the existing custom-endpoint pattern. Happy to open a PR.
(Context: I'm the developer of AnyLLM — happy to provide a sandbox key for testing.)
Best,
Leo Bennett
leo.indiedev@gmail.com