News

Dashboard, assistant list and files of any type

New

• Dashboard — /en/dashboard. Paste your JWT to see the token's expiry date, your limits, the available assistants and their models, your tasks, and a playground where you can send a task without writing code. Every assistant comes with a ready-to-run request example in curl, JavaScript and Python.
• GET /api/me — a token check: who owns it, when it expires, the active task limit and how much of it is in use.
• GET /api/assistants and GET /api/assistants/{slug} — the assistants available to the token and their models: where to get assistant_id and llm_model_id. More in Assistants.
• Live chat on the home page — paste your JWT and pick an assistant from the list, no need to type an ID.
• SDK 0.4.0 — getMe(), listAssistants() and getAssistant(slug) methods.

Changed

• POST /api/files accepts files of any type: archives, office documents, source code, audio and video. If a file with a text extension turns out not to be text, the response is a regular 422 with errors.file.
• Video is generated right inside the task and arrives as a file item in output — just like images.
• Turn cost is no longer disclosed: metadata.cost, tool_call.cost and messages[].usage.provider_cost are now always 0. The fields are kept so code that reads them doesn't break.
• Context window overflow is now a clear task error: how many tokens piled up, how many fit, and what to do — drop older turns from turns. The history is never compacted automatically.
• Assistants called through Agent don't run code for now. Reading spreadsheets and documents works as usual.
• The site and docs no longer load fonts or styles from third-party CDNs.

Fixed

• The assistant can see files produced in previous tasks when you pass them in the turns history.
• File upload errors now come back as clear JSON instead of the raw server response.