Loading...
Loading...
Run OpenAI Codex CLI on AisaysAI models. We implement the Responses API natively, so Codex connects with its default wire protocol — no patches or proxy layers required.
sk-or-v1-).~/.codex/config.toml:# ~/.codex/config.toml
model_provider = "aisays"
model = "zhipu/glm-5.3"
[model_providers.aisays]
name = "Aisays.ai"
base_url = "https://api.aisays.ai/v1"
env_key = "AISAYS_API_KEY"
wire_api = "responses"export AISAYS_API_KEY="sk-or-v1-..."
codexPrefer the chat wire API? Swap wire_api — everything else stays the same:
# Alternative: use the Chat Completions wire API instead
[model_providers.aisays]
name = "Aisays.ai"
base_url = "https://api.aisays.ai/v1"
env_key = "AISAYS_API_KEY"
wire_api = "chat"| Feature | Status |
|---|---|
| wire_api = "responses" | Fully Supported (native — recommended) |
| wire_api = "chat" | Fully Supported |
| instructions | Fully Supported → system prompt |
| input (message items) | Fully Supported (text) |
| input (function_call / function_call_output) | Fully Supported — tool round-trips work |
| tools (function, strict) | Fully Supported, incl. parallel calls |
| stream | Fully Supported (response.* SSE events) |
| usage / billing | Real token counts reported every request |
| reasoning | Consumed upstream; not returned to Codex |
| store | Stateless — Codex sends full history each turn |
| web_search tool | Not Supported (rejected with a clear error) |
| images / input_image | Not Supported (rejected with a clear error) |
Pick a model from the model catalog. Also see the Claude Code guide.