Skip to content
- Every guide under Agents, Coding tools and SDKs links the tool’s official site and shows how to
install it.
- Each of those guides has a Get API key button, so you can grab a key without leaving the page.
- Links to other sites now open in a new tab, so you keep your place in the docs.
- A new Prompt to your agent button on the integration guides, the Quickstart, Authentication, Text
generation, Embeddings, Images and Video copies setup instructions you can paste into a coding
agent.
- The home page no longer lists recommended text models. The Models page has the full catalog.
- The Bayle chat now works as a general assistant. It answers questions about Bayle when you ask,
instead of steering every conversation toward it.
- In the sidebar, a chat’s delete button now shows only when you point at it, and the whole sidebar
scrolls as one area, with the menu entries and your chats moving together.
- Settings and a new Support option (help on WhatsApp in one tap) now sit in one menu at the bottom
of the sidebar.
- On phones, the menu now opens as a side panel that slides in from the left.
- The Models page has the same privacy, restriction and sort filters as the model picker.
- The API page’s code examples are now highlighted and neatly formatted.
reasoning_effort is documented on POST /chat/completions, along with each model’s default
thinking level.- Chat completions carry no hidden system message. Your messages reach the model exactly as you
sent them.
- Long chats can be summarized automatically, so a conversation keeps going past the model’s
context window. Turn it off per chat with the auto-summary toggle.
- Privacy now has two tiers, Private and Anonymous, shown with the same shields everywhere.
- New API page: first steps, setup for coding agents, and a daily usage chart per model with CSV
export.
- View all models opens the Models page.
- Every chat has its own link at
chat.bayle.ai/<id>. Back and Forward move between chats, and
the browser tab shows the chat’s title.
- The model picker reads better on phones: names can take two lines, and the filter and sort
menus stay anchored to the search row.
- Text model prices are shown per 100K tokens, and context sizes read 32k, 128k or 1M.
- The composer shows how full the current context window is.
- The picker’s tab counts follow your search and filters.
- Reasoning streams live into the thinking row, and the thinking timer keeps running when you
switch chats.
- Brief failures are retried inside the same answer, keeping the text already streamed. When a
retry cannot help, the message says why.
- An unsent image or video prompt is kept when you navigate away and back.
- Every model shows its privacy shield and restriction gauge, in the picker and wherever a model
is named.
- Code blocks are editable, even while the answer is still streaming. Your edit survives a reload
and is sent with your next message.
- The HTML viewer has a Copy button and an editable Code pane.
- An unsent message survives a reload of the tab.
- HTML in an answer runs in a sandboxed viewer, with a Fix action that repairs a broken page in
place.
- Long chat answers are no longer cut off after five minutes.
- New look on the Bayle brand. Code blocks are cards with a sticky header and a download button.
- The model picker sorts by price or by name.
POST /images/generations documents its size options and moderation. quality is accepted
and ignored.
- A request that runs out of time answers
request_timeout, distinct from a client that
disconnected.
- Image generation has its own time limit and returns a coded
504 when it runs out.
- Image and video generations live in the same conversation as your messages, each labelled with
the model that made it.
- Enter submits a generation prompt.
POST /video/queue takes end_image_url and negative_prompt.
- Video jobs show as cards that track their progress and survive a reload.
- Chat completions accept file content parts.
- Every route answers cross-origin requests, so you can call the API from a browser.
- Video generation:
POST /video/quote, /video/queue, /video/retrieve and /video/complete.
- The API launches with
GET /models, POST /chat/completions (streaming, tools and vision),
POST /embeddings and POST /images/generations.