... just shipped Projects in Osaurus and it fixes one of the oldest annoyances with AI chat apps: starting from zero every time. Group related chats into a project and they share three things: instructions (one set of context for every chat), knowledge (document collections any chat can search, with per-agent access you approve), and memory (a fact learned in one chat is recalled by every agent in the project). The part I like most: memory survives across agents ... ... answers correctly. Nothing re-explained. Still local, still free, still MIT licensed. Download at osaurus.ai and tell me what breaks. We prefer negative feedback over positive. Comment from Maria McGinn(@mariabuildsbriskly): @chrismessina  I did laugh at that @Osaurus

... Everyone talks about model lock-in, but the real switching cost is your conversation history. Months of context, decisions, and half-finished projects living inside one app. We just shipped chat import in Osaurus (free, open source Mac app) so you can export your conversations from ChatGPT, Claude, Grok, Gemini, or Open WebUI and continue them with any model. Building it made us realize how differently each app treats your data on the way out. Some exports are clean JSON ...

... Osaurus agents can now drive a real browser. You give the agent a goal in plain language, and it navigates, clicks, reads pages, and reports back. The demo we shipped it with: a 9B model checking an Amazon shopping cart. Real browser, real login, real cart. It signed in, read the page, and returned the items and prices. About 90 seconds, running locally on a Mac. What makes it different from cloud browser agents: The model runs on your machine ... ... passes through an approval gate: reading and navigating are automatic, but typing, submitting forms, and anything that looks like a purchase asks you first. Sign-ins happen in a visible window you complete yourself. The model never touches credentials. Osaurus

... Quick follow-up from the Osaurus team. On Tuesday, PrismML released Bonsai 27B, a big open source AI model compressed to a fraction of its normal size. A model like this usually needs 54 GB of memory, which means expensive cloud servers or a maxed-out workstation. This one fits in under 8 GB. That's small enough for a regular MacBook. Why this matters: the models keep getting smaller while staying smart. Every time that happens, "AI that runs ... ... noticeably worse AI. That gap is closing fast. The catch was that the standard tools on Mac can't actually run models compressed this way yet. So we did the conversion work ourselves and shipped it. If you tried Osaurus

... Just shipped Knowledge Base in Osaurus. You hand your agents a folder of reference material, itineraries, specs, spreadsheets, whatever you keep, and they search across all of it and answer from it. It runs on your Mac. The decision we kept coming back to: search alone wasn't enough. An agent that reads your files and gives a confident answer is still a black box unless it shows its work. So every answer cites the exact source it pulled from ...