What is new in StickyPrompts.
New capabilities, controls and fixes, newest first. New models land in the model picker as providers ship them, without a migration on your side.
-
Compact a long conversation
Long chats no longer have to end when the model's context window fills up. Compact one and keep working in the same chat: the AI summarizes the conversation so far and carries on from the summary and your newer messages.
- New Compact conversation: the icon under the latest reply, or Compact in Chat info, summarizes the chat so far; a Conversation compacted divider marks the point and opens to show the summary
- New Automatic compaction: a chat is compacted on its own once it reaches 80% of the model's context window. Change the threshold, set a token count or switch it off in Settings, Preferences, Conversation compaction
- Improved Chat info shows Context: how much of the model's context window the chat uses, next to the size of that window
- Improved Earlier messages stay in the chat after a compaction, so you can still read and copy them
- New REST API: POST /conversation/:id/compact compacts a conversation, and a fork with compact: true is compacted straight away
-
Decisions, powered by Jev
A new app under Apps: give it a piece of text and a few typed questions, and Jev answers each one with one of your options, a level on your scale or a true / false value, with its confidence.
- New Decisions in Apps: paste any text, such as a support ticket or a review, add Choice, Score and True / false questions, and run them with Jev 1.13 from TypeSafe AI
- New Every answer is typed: the chosen option, a level on your scale or a value from 0 to 1, with a confidence and each option's probability
- Improved Decisions are saved, and your workspace's guardrails, model access and data residency rules apply to them
- New REST API: POST /decision runs one and returns its answers, GET /decision lists them (API key permissions decision_get, decision_manage and decision_delete)
-
Claude Haiku 5.5
Anthropic's new small model, released on 7 October, is available in every workspace that allows Anthropic models: in chat, agents, prompts, workflows and Compare Models.
- New models Claude Haiku 5.5, Anthropic's fast model for high-volume work such as classification, extraction, routing and subagent tasks, with a 1M token context window
- Improved Set it as your Preferred AI model in Settings, Preferences, or compare it with Sonnet 5.5 side by side
-
Snippets, and response styles for the whole workspace
Tell the AI who you are, what your company does and how your brand sounds, once. Snippets are reusable pieces of context you switch on in any chat, several at once, and response styles can now be shared with everyone in the workspace.
- New Snippets: the new puzzle button in the composer lists your snippets and the workspace's; tick up to 20 and select Apply, and they stay on for that chat
- New Create snippets in Settings, Snippets: a name, a description and up to 20,000 characters of content, for Only me or the Whole workspace
- New Workspace response styles: share a style with everyone, and it appears under Workspace in every member's Style picker
- Improved New AI Behaviour tab in Workspace settings brings the workspace prompt, workspace snippets and workspace response styles together
- Improved The Style picker has a search box and groups styles into My styles, Workspace and Built-in
-
Claude Sonnet 5.5
Anthropic's new Sonnet, released on 28 September, is available in every workspace that allows Anthropic models: in chat, agents, prompts, workflows and Compare Models.
- New models Claude Sonnet 5.5, Anthropic's faster companion to Claude Opus 5.5 for well-scoped everyday work
- Improved Set it as your Preferred AI model in Settings, Preferences, or compare it with Opus 5.5 side by side
-
Team chats
A whole team can now work in one conversation. People talk to each other as usual, and the AI answers only when someone asks it to, with the full thread as context.
- New Pick a team in the sidebar's TEAM switcher and start a team chat, or use the Teams section of the New chat menu
- New The AI replies only when someone mentions @ai or turns on the @AI button in the composer
- New Type @ to bring in the AI or an agent that has been invited to the chat
- Improved The AI's answer stays in the thread, visible to the whole team
-
EU data residency, end to end
Workspace data is stored in the EU by default, and admins can now require that inference stays in the EU too, including for open-weight models.
- New Data Residency settings: EU storage by default, with US hosting on request for Enterprise
- New Require EU-served inference endpoints, so open-weight models run from EU hosts wherever the lab that made them is based
- New Zero data retention only: restrict the catalogue to providers that do not keep your data
- Improved See the impact on your model catalogue live, before you save; models that cannot comply are greyed out
-
SSH bridge for private databases
Connect a database that has no public endpoint by going through an SSH jump host you control.
- New SSH bridge for PostgreSQL, MySQL, MariaDB and SQL Server connections
- Security Key-based: add the public key StickyPrompts shows you to the SSH user's authorized_keys
- Security Only StickyPrompts' single IP address needs to be allowed through to the jump host
- Improved Permission levels per connection: Read only, Read & write or Full access
Everything here arrived without a migration.
New capabilities arrive inside the guardrails and model access you already set. Start free with $5 of AI usage.