Changelog
What's new in Context Engine
Every Context Engine release, newest first, and what got better in it.
Get a mail when a new version is out
One short mail per release with what's new. Unsubscribe from any of them with one click.
Fixed
- The day's journal is no longer thrown away when the assistant lists the day's moments out of order or lists too many; it puts them in order and combines the closest ones, keeping everything that was written.
- More of the examples your AI tools get for saving to memory can now be saved as they are.
- When something sent to your engine is rejected, the error is now short and to the point, even for very large requests.
Improved
- When an AI tool saves something to memory in bulk and one item is rejected, the reply now says which field is wrong and why.
- A type you describe when you add it to your memory now shows that description to every AI tool that asks what the type is for.
Fixed
- The examples your AI tools get for the kinds of memory you added yourselves can now be saved as they are, instead of being rejected.
- Asking for a version of the memory format that doesn't exist now gets a clear answer listing the versions that do.
Fixed
- Text in angle brackets, such as
Promise<Response>, now shows as written instead of disappearing. - A line break inside a table cell now shows as a line break.
- Google Chat keeps working after its app moves to a different Google Cloud project, and subscribing a space again no longer fails.
- The health status no longer reports the work queue as stuck when a job is simply scheduled for later.
- Saving to memory through the AI tools no longer fails when some of its fields are left empty.
Improved
- Chat now tells you when Claude's usage limit resets instead of showing "thinking…", in Telegram, Slack, Google Chat and WhatsApp.
- Journals, Flows and scheduled jobs that reach Claude's usage limit now wait and try again instead of failing.
- This release includes security updates to the software the engine is built on.
Fixed
- A week in which nothing happened no longer makes the health status report a problem with journals.
- A weekly journal no longer sometimes shows as a grey block of code.
- Entries with very long names now show up in search.
New
- A day whose journal couldn't be written because Claude was unavailable, for example after the sign-in expired, now gets its journal automatically once Claude works again, oldest day first.
Fixed
- A journal run that stopped part-way through is now marked as failed, instead of looking as if it were still running.
Improved
- The engine's health status now shows when the engine has been signed out of Claude, so you find out before journals and answers stop.
Fixed
- When the Claude sign-in has expired, work now stops at once with one clear message, instead of retrying for a long time as if Claude were busy.
Improved
- When one of the engine's AI connections is down, the engine now says why.
Fixed
- When the Claude version on your engine is too old for the selected model, the assistant now says so on the first message and explains how to fix it. Before, chat rooms kept showing "thinking…" and never answered.
- A single slow reply from OpenAI no longer makes the engine's health status report a problem with its AI connections.
- A quiet day with little new to write about no longer makes the health status report a problem with journals.
Improved
- When a flow can't answer, for example because the Claude sign-in has expired or a spending limit was reached, the reason now appears in the flow's room and stays there.
- Everyone who manages a flow now sees the conversation in its room as it happens, not only the person who started it.
- When Claude is busy, a flow's room now waits and answers instead of asking you to try again.
Fixed
- Asking a flow to rework a long instruction no longer fails with "Could not send". Your question is saved straight away, and the answer appears in the room as soon as it is ready.