Fostera v0.7.0 — Claude Fable 5.1 & chat loading fixes

New Features

• Added Claude Fable 5.1, Anthropic's most capable model, for long-running agentic work and demanding reasoning. It replaces Fable 5 as the auto-routed option; Fable 5 stays hand-selectable.

Improvements

• A chat that fails to load now tells you what actually happened. Being signed out, a chat that no longer exists, being offline, and a fault on our side each say so plainly instead of all reading "check your connection" — and where retrying cannot help, you get the action that can.
• Your draft stays yours while a chat is loading. The composer used to grey out and lock the text you had already typed at exactly the moment you most wanted to copy it; you can now keep typing, and only sending waits.
• Retrying a chat that did not load shows that it is working, and one dropped request no longer strands the whole conversation — it retries itself.
• Private pages are no longer kept in shared browser storage. Account-scoped pages were being saved in a cache that is per-browser rather than per-account, in a form that survived signing out.
• Updates moved into the main navigation on the public site, instead of sitting behind the overflow menu. Navigation labels and links now come from one shared list, so the desktop bar, the mobile menu and the More menu can no longer disagree about where something lives.
• An outage in the service that paces requests no longer takes chat down with it. It sits in front of every chat read and send, and when it was unreachable it failed the whole request; it now falls back to pacing locally so chat keeps working.

Bug Fixes

• Fixed chats failing to open at all. An image library used by document and image generation was being loaded on the path that simply READS a conversation, and a missing piece of it made every one of those reads fail. This is the "This chat couldn't load" and "I cannot get into my first time chat" people wrote in about.
• Fixed Souls denying things they can actually do. Asked what it could do with a connected Gmail, a Soul replied that it could not send, reply, or act on your behalf — but drafting, sending and replying are shipped abilities that simply wait for your approval. A Soul only sees the tools loaded for the turn it is answering, and that set is deliberately narrower than what you have enabled; it now knows the difference between "not loaded right now" and "cannot do".
• Fixed a Soul appearing to say "Something went wrong. Try sending your message again." Platform errors were being rendered inside the Soul's own message bubble, typed out character by character in its voice; they now appear as what they are, with a real explanation of what failed.
• Fixed Souls starting in learning mode after every update. Finishing a Soul's first edit did not always get recorded, so every launch asked again — permanently, for anyone whose confirmation was lost. It is now remembered and retried until it lands.
• Fixed error messages running together into one unreadable line ("This chat couldn't loadYour conversation is safe."). The heading and body were being rendered as the same line of text; the same fault affected 41 places across the app.
• The Resonance pill beside the composer no longer carries a small monthly bar. It sat misaligned with the balance next to it, and it was measuring only your monthly allowance while the number beside it was the total of your monthly and wallet balances — so the two read as a ratio they were never in. The full breakdown is still one tap away.
• When a Soul falls back to a second model, you are now charged once, for the model that actually answered. Previously the charge landed on the first model tried even when it failed, and the model that did answer went unrecorded entirely.