AeionClaw + OpenAI / ChatGPT
Configure OpenAI (GPT-5, GPT-5 Mini, o3 reasoning) or Azure OpenAI as AeionClaw's LLM provider. Strong general-purpose model with the largest ecosystem of fine-tuning + function-calling tooling. Per-tenant API key, no-training endpoint, Azure-native option, transparent cost attribution.
OpenAI Model Matrix
`
Setup Walkthrough
```bash
Azure OpenAI — Same API, Enterprise Procurement
`
Function Calling — Parallel Tool Dispatch
`
FAQ
No — AeionClaw uses the Chat Completions API directly. AeionClaw's own conversation + memory + tool architecture replaces the Assistants API equivalent. Configure as if it's any OpenAI deployment.
A fine-tuned model shows up under its own deployment/model ID from OpenAI, and can be entered as the `model` on a request like any other OpenAI model ID — AeionClaw doesn't have a separate fine-tuning management UI. Note: training data should be reviewed for PII before fine-tuning.
ChatGPT Enterprise is the consumer-facing product. AeionClaw uses the OpenAI API directly. Different products, different procurement — but the underlying models are the same family.
Use for: complex math (financial reconciliation), multi-step debugging (incident root-cause), code generation (custom skills), planning (workflow composition). Skip for: high-volume + low-stakes (use GPT-5 Mini instead) or low-latency-required tasks (reasoning adds several seconds).
Yes — GPT-5 handles images natively via the OpenAI adapter (image_url content blocks in the request). AeionClaw can pass uploaded PDFs, receipts, screenshots, diagrams for analysis. Per-image token cost is on top of text.
AeionClaw's OpenAI adapter can call Whisper (`whisper-1`) directly for transcription. Aeion Connect uses this for call transcription in some configurations. Per-minute audio cost ~$0.006 via the Whisper API.
OpenAI's API offers a DPA (Data Processing Agreement) and doesn't train on API data by default. For strict EU-residency requirements, Azure OpenAI's EU regions are the closer fit since Azure resources can be pinned to a specific region.
OpenAI's rate limits scale with usage tier. AeionClaw doesn't currently queue bursts or auto-failover to a different provider on a hard rate-limit response — a rate-limited call fails and needs to be retried by the caller.