Agently already lists every major Chinese provider as an OpenAI-compatible option — DeepSeek, Qwen, Kimi, GLM, MiniMax, Doubao, SiliconFlow, ERNIE — which is exactly the right instinct for a global audience. The problem isn't coverage on your end; it's that international users hit a wall at signup: Chinese phone number, Alipay, language barriers — for each provider individually. That gap between "we support these providers" and "users outside China can actually use them" is where things stall.
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 Agently
Model pool, one endpoint. Agently's Model Pool pattern (activate_model() across multiple configs) is designed for exactly this — but maintaining separate base_url / api_key blocks for DeepSeek, Qwen, GLM, Kimi, MiniMax, Doubao, ERNIE is seven signups and seven config entries. AnyLLM collapses them into one connection, and activate_model() still switches the model name — the routing still happens at your layer.
Works today, zero integration. settings.yaml with ${ENV.OPENAI_BASE_URL} and ${ENV.OPENAI_API_KEY} — point both at AnyLLM and you're done. No new plugin, no adapter code.
Key pool × failover, doubled. Agently's API key pool rotates on 401/403/429; AnyLLM adds provider-level failover beneath that. Two independent resilience layers, no overlap.
TriggerFlow & Dynamic Task are token-heavy. Signal-driven branching, fan-out, DAG execution — these multiply LLM calls per task. Pass-through pricing with zero markup on Chinese models makes larger workflows practical.
Default example already uses DeepSeek. Your quickstart uses deepseek-v4-flash — that's the exact tier AnyLLM serves natively, plus the rest of the Chinese frontier lineup in the same key.
Proposal
1.A docs section: "Using an OpenAI-compatible gateway (AnyLLM)" — showing the collapsed settings.yaml for international users who want the full Chinese model pool without per-provider signup.
2.Happy to open a PR with the docs example, or provide a sandbox key for the team to test activate_model() across Chinese models.
(Context: I'm the developer of AnyLLM.)
Best,
Leo Bennett
leo.indiedev@gmail.com
Agently already lists every major Chinese provider as an OpenAI-compatible option — DeepSeek, Qwen, Kimi, GLM, MiniMax, Doubao, SiliconFlow, ERNIE — which is exactly the right instinct for a global audience. The problem isn't coverage on your end; it's that international users hit a wall at signup: Chinese phone number, Alipay, language barriers — for each provider individually. That gap between "we support these providers" and "users outside China can actually use them" is where things stall.
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 Agently
Model pool, one endpoint. Agently's Model Pool pattern (activate_model() across multiple configs) is designed for exactly this — but maintaining separate base_url / api_key blocks for DeepSeek, Qwen, GLM, Kimi, MiniMax, Doubao, ERNIE is seven signups and seven config entries. AnyLLM collapses them into one connection, and activate_model() still switches the model name — the routing still happens at your layer.
Works today, zero integration. settings.yaml with ${ENV.OPENAI_BASE_URL} and ${ENV.OPENAI_API_KEY} — point both at AnyLLM and you're done. No new plugin, no adapter code.
Key pool × failover, doubled. Agently's API key pool rotates on 401/403/429; AnyLLM adds provider-level failover beneath that. Two independent resilience layers, no overlap.
TriggerFlow & Dynamic Task are token-heavy. Signal-driven branching, fan-out, DAG execution — these multiply LLM calls per task. Pass-through pricing with zero markup on Chinese models makes larger workflows practical.
Default example already uses DeepSeek. Your quickstart uses deepseek-v4-flash — that's the exact tier AnyLLM serves natively, plus the rest of the Chinese frontier lineup in the same key.
Proposal
1.A docs section: "Using an OpenAI-compatible gateway (AnyLLM)" — showing the collapsed settings.yaml for international users who want the full Chinese model pool without per-provider signup.
2.Happy to open a PR with the docs example, or provide a sandbox key for the team to test activate_model() across Chinese models.
(Context: I'm the developer of AnyLLM.)
Best,
Leo Bennett
leo.indiedev@gmail.com