Public AI / REPORT 001
TH AI Passport recap: developers need more than a chat page
A two-hour hands-on review argues that a public AI platform needs APIs, harnesses, and practical developer tools.
Editorial update: the original post was written before the platform’s API documentation was found. OKMD AI Playground was later confirmed to offer an OpenAI-compatible API. The following text is preserved as a record of the TH AI Passport test at that time.
The review reflects roughly two hours of use and a perspective shaped by sustained work with GPT, Gemini, Claude Code, and Codex.
A model catalogue is not a working environment
Putting many models in one website improves access, but developers gain much more when those models connect to CLI tools, IDEs, and agent harnesses through an API or compatible proxy.
A strong model without web access, files, code execution, and result verification cannot use its full capability. The harness provides those tools and the context required to work.
Why coding agents differ from chat
Claude Code, Codex, and similar agents can inspect a project, edit files, run commands, test the result, and continue across several steps.
That is why the same model can perform very differently across products. Developers pay for the execution and verification loop, not simply for another chat response.
Model Council as a middle path
If full API or tool access is too expensive, a Model Council could assign planning, implementation, and review to different models.
This would give a multi-model platform value beyond a drop-down list of providers.
Keep pace with a fast market
The model market changes daily. A public platform should offer strong current models from several regions and explain each model’s limits clearly.
If the product remains only a chat hub, developers who could produce deeper use cases may still avoid it, even when access is free.
SOURCES / 02