AgentOne 1.1.2 is a "point release" in name only. It has one of the most requested features yet, a rebuilt settings page, a new first-run experience, 60 new providers, and a big round of performance work.
Long Chats Can Keep Going with Compaction
Every AI model has a limited context window, which is how much of a conversation it can see at once. Until now, a long enough chat would eventually hit that limit and stop working.
AgentOne can now compact the conversation. It writes a detailed summary of everything so far, then continues from that summary with a fresh context window. Auto-Compact is on by default and kicks in when a chat uses 90% of the model's context window. It works in the middle of a response too, so long tasks with lots of tool calls keep running instead of failing partway through.
You're still in control. Open the model configuration next to the model selector to turn Auto-Compact off, change the threshold (anywhere from 5% to 95%), or click Compact now. Compacted chats show a "Conversation compacted" divider, and clicking it shows the summary AgentOne is working from. Your earlier messages stay visible in the chat; only the AI's view of them is replaced. If you edit or regenerate a message from before the divider, the compaction is undone and the full history comes back.
AgentOne Can Read Your Past Chats
The new Chat History tools let AgentOne search your past conversations and read through them. Ask "What did we decide about the trip last week?" or "Pick up where we left off in the database chat," and AgentOne can find the chat and read the parts it needs.
Both tools ask for your approval by default, and the approval card links to the chat AgentOne wants to read. Dragging a chat from the sidebar into the message box now attaches a lightweight reference instead of the whole conversation, which keeps the context window free for your actual work.
These tools also back up Compaction: if AgentOne can't write a summary during a response, it uses Chat History to read back through the chat and pick up where it left off.
Settings, Rebuilt and Searchable
Settings is now one scrollable page with an "On this page" sidebar that follows along as you scroll. Press Cmd/Ctrl + K to search every setting. Search understands synonyms, so "dark mode," "night mode," and "color scheme" all find the Theme setting.
Provider settings now list every built-in model (not just the ones you've added), and you can override any model's details right from the list. The Account section is clearer about monthly credits, too.
A New First-Run Experience
Onboarding has been redesigned so signing in comes first, and if you'd rather bring your own API keys, that option is right there. New users also get a quick guided tour of the message box, model selector, chats, and account menu. You can take it again anytime from Help & Updates.
Everything Else
There's a lot more in this release:
- 60 new providers, bringing AgentOne to 127 built-in providers, including Nebius Token Factory, Meta, Z.ai, Xiaomi MiMo, and Tinfoil.
- [Beta] Reasoning Effort, so you can choose how much a model thinks before answering, from None to Max.
- Using your own Anthropic API key works reliably again, along with a fix for providers like LM Studio that rejected requests.
- A new account menu at the bottom of the sidebar, and clickable notifications that take you straight to the chat they're about.
- Custom fonts, plus Sky as the new default primary color.
- Subagents start fresh with a self-contained task instead of the whole parent conversation, which saves tokens and keeps them focused.
- Scheduled agents skip runs that would start more than 5 minutes late, and no longer flash a PowerShell window on Windows.
On top of that, there's a large round of performance work, especially for long chats and streaming responses. Saving a chat no longer re-renders most of the app, streaming Markdown only re-processes the part that's still changing, and the model catalog uses much less memory.
Get It
AgentOne 1.1.2 is out now. If you already have AgentOne, it should update automatically. If not, you can download it for free.
As always, I'd love to hear what you think on Discord or the forum.
Top comments (1)
how does the new compaction process interact with the existing chat history storage, are you using a background vacuum or a more aggressive rewrite, and what impact does that have on read latency for the searchable set?