Changelog

v1.9

Your Own AI 0.8.0 - your own server as your hardware, the working session, tools that look like chat

What changed in these docs, page by page:

• Routing modes loses the Auto - My Hardware row. A new section, "Your own server is your own hardware", says that a server you run counts as your offline models in both Auto modes, takes a question when it holds the clearly stronger model, says "your server" on the reply, and is set aside for a minute when it stops answering; an AI still set to the old choice behaves as Offline Only.
• How routing decides adds the server to "On your device" and states the quality-first pick for project and tools sessions: the most capable model that runs on this computer, by capability, then size, then fit, one reload per session.
• Routing settings rewrites "Working on projects" around that pick and adds the "Project work runs on" choice (Auto, this computer, your server), with the reading-room rule that keeps work here when the server's is smaller.
• Attachments and images drops My Hardware from the files rule.
• Engines and your hardware rewrites "Your own server" (no separate mode, the card's window, speed and models, a served turn says so, Fine-tune does not reach a server), adds "Several graphics cards" (the biggest card alone unless the model needs pooling; grades against the biggest card), and rewrites "Utility work stays small": the helpers use the graphics card when there is room, Automatic or Keep on the processor, and give memory back under pressure.
• Fine-tune now opens on what the word means (how a model runs on this computer), adds what Measure on this computer also decides (batches, the integrated-graphics check, the Auto stop), the Graphics cards row, and moves the four writing sliders to a "Reply style" section: Reply style for every AI in Settings, Reply style for this AI in the AI's form. Your AIs renames its section to match.
• The helper says where the two models run now: on the graphics card when there is room after the chat model, otherwise the processor.
• Offline models adds Hemmingway 1 as a licensed model (free for personal use, not for business use) and a new section, "When this computer cannot run a model": the one card, the offline fix first, the single line about online models, and how grades are sized.
• The working session is rewritten: steps named after what they are, glyphs and status colors, thoughts as rows, the plan count, Simple or Detailed (projects only, Ctrl+O, Esc stops), the header summary, background tasks and helpers with a Stop and "Stopped", the permission card above the input, fast to the first step, saying something while it works, when a small model struggles, more of the truth in the record, the desktop notification, and waiting on your server.
• Projects and Permissions and approvals follow: Auto asks with a card instead of refusing, the card waits above the input field and the rail keeps a receipt; Good to know gains the quality-first pick, the server, the struggle note, interjecting and the notification.
• Tools is rewritten around the app's Tools page: Your Tools and Available Tools, one Set up list per tool with a button per line, the Obsidian and Logseq cards, settings as text, switches and folders with a real Save, tools only when a message needs them and "Answer again with tools", the tools chip opening a tool's Set up list, and "What a tools turn looks like" as chat with tools now is - the page changes on the keystroke, collapsed rows while it runs, one grey line when done, "Answered from its documents" or "Answered without using its tools" with Try again with the tool. "What tools need from your model" carries the quality-first pick and the struggle note.
• Skills: a skill can carry its own icon.
• Your documents gains "Keep a folder in sync" (Obsidian and Logseq notes, cloud-drive files left alone, the folder decides inside a synced folder) and "Documents stay linked to their files" (Check again, Relink, Remove, Locate, Search for them, one job at a time with a Stop, selecting several), says that only passages that fit the question are given and that follow-ups are matched with the question before them, describes the Sources button, and notes read as notes. The backup section carries synced folders.
• Backups and recovery, Your records, Each AI's memory (search inside conversations), Your first chat (say something while it answers) and Help and diagnostics (what the report and the matrix's speed figures say) each gain a paragraph.

v1.8

Your Own AI 0.7.2 - the activity card, the fine-tune dialog, deletions across devices

• Fine-tune is rewritten for the redesigned dialog: what the model runs at now and where that came from, Measure on this computer as the main event with its cost stated, the Faster answers to More room slider with pages in view and words a second, and the Set it yourself drawer.
• Your documents and Welcome describe the activity card: every download and every document being read, on every page, with the pieces inside the current document, a Stop, and the AI it is for.
• Backups and recovery: a conversation deleted on any of your computers stays deleted after a restore, since each backup carries what its computer deleted.

v1.7

Your Own AI 0.7.1 - your documents, backups that keep up

• Your documents is a new page under Memory: one collection of your own documents, yours rather than any one AI's, read whole on your device with progress and a Stop button, folders included. Each document gets a card written on your device, a Mine tag guessed from the file's own author field, and the author beside its name. Every question hands the AI a card for each document it has and the passages that match, sized to the room a local model really has, and answers cite the document by name.
• Your Memory describes the block "What your AIs are told about you": the picture written from your facts and the paragraph written from your documents, shown on request, and exactly where that text goes.
• Backups and recovery gains "When backups run": after a chat, a few minutes after launch, daily, and the Back up now button; a locked Vault is waited for; only what changed is read; the Vault is checked against what the device believes it stored. The restore section says what a restore never does (double, or bring back a deletion) and how documents are read back in from where they were.
• Knowledge and packs and Each AI's memory point documents at the new page; authored text entries stay where they were.
• Fine-tune gains the Context memory setting (Standard, Compact, Auto) and what Tune on this computer teaches the automatics: the compact cache and the measured expert split.
• Each AI's memory describes finding conversations by title, narrowing by source, and deleting many at once.
• Skills updates cover listings as well as links, and the Signed by @username card. Your first chat notes the Continue link and its switch under Appearance.

v1.6

Your Own AI 0.7.0 - routing around one dial, the welcome, and the helper

• Routing is rewritten around how 0.7.0 decides: an Online and Offline AI answers with a frontier online model by default and keeps a question on your device when a model here is better for it. One dial - Frontier-first, Balanced, Local-first - replaces the old freshness setting, and the online models are set by job: Everyday, Hard, Live web, Agent, Planning, with the recommended names coming from the router.
• Thinking and follow-ups: hard questions think before answering on a capable model, a short follow-up stays with the thread's model, and the feedback loop (Redo on your device, Try this answer online) and its reset are described.
• Vision: a model's vision add-on loads only when you send an image, so a vision-capable model is graded for text alone.
• Getting Started gains the welcome: three steps, one model picked for your computer, the Personal or Work set of three AIs, and a download that never blocks.
• Models gains the helper model - memory, recall, routing, documents - offered from the home page on the second launch, private and removable.
• Fine-tune: per-model context, expert split and speed-up file; Tune on this computer with the Speed-Room slider; Settings › Engines "Fine-tune this computer"; the per-AI advanced generation settings and their inheritance order.
• Tools in chat: the chip beside the model chip, tools and skills on or off, per-AI approvals, the reading-room reload, and plain-spoken turns.
• Projects: checks after edits, Undo this turn's changes, and the agent's fallback for a picture its model cannot see.
• Online models state the billing rule: the card is charged when accrued overage reaches $100 and at month end, and online pauses past your allowance until an invoice is paid.
• Your Data tells the Flowsta Vault story - what it is, what it does here, what it takes - in Backups and recovery.
• Help: a new section covers the diagnostic report and the truth matrix.
• Wording: "Flowsta identity" everywhere, exports as Share publicly or Export to file, characters at Add-ons › Characters, and records described as sealed on your device.
• Navigation: Add-ons has its own top-level entry and sidebar, and Models lists the helper model and Fine-tune pages.
• Sharing: "Claiming a listing that names you" - a skill or tool listed by Flowsta from a public source can be claimed by its maker with one line in the source, linking it to their Flowsta username and following their repository from then on.

v1.5

Add-ons gets its own section

Six pages for the Add-ons menu as it ships with tools: the four kinds and the split between what is for your AIs (characters, skills, tools - chosen per AI) and what is for this computer (components - got in Add-ons, managed in Settings › Storage); Characters as complete signed packs with Add to Your AIs; Skills in the Agent Skills open standard, carried into chat by meaning and into projects on demand; Tools as MCP servers with what they need installed from the app, the settings they ask for kept on the device, guidance that rides into every session, and the same approvals in projects and in chat; Components; and Sharing - one button, signed with a Flowsta identity, what goes out and what never does, how checks, the AI review and a person decide, and the status line you see in the app afterwards.

v1.4

Routing that judges how a model runs, and the rules in one place

The routing pages describe routing as it ships in Your Own AI 0.6.0: picks are about how a model runs on your machine - a mixture-of-experts model running GPU + RAM counts as running well, a dense model that does not fit the card is too large - your own measured speeds and load times count, and a question bigger than a model's reading room goes to one that can hold it. The rules live here now; Settings → Routing links to this guide and keeps the data - how routing sees your models and its recent decisions. Projects: how side-work and whole sessions route, the readiness bar for local agent work, and the nudge when a local model is as capable. Models documents the GPU + RAM badge and the expert-lean builds, multi-part downloads for very large models, speed-up files, measured speed on the model list, and damaged-file handling, where models are stored (the movable folder), disk-checked downloads, and the active-experts check behind the GPU + RAM promise. Memory is rewritten around two kinds - each AI's own memory, and Your Memory shared by all - plus the on-device portrait: how your AIs see you, written and shown on your device, never on a server.

Memory gets its own section

Memory moves out of Your AIs to a top-level section of five pages: why ours is different (the whole loop on your device, shown to you), Your Memory with the "How your AIs see you" summary and the fix-the-fact rule, each AI's memory, project notes, and what survives your laptop. The old address redirects.

v1.3

Your AIs, Routing, and the product as it is now

The docs are reorganized around how the app works today. "Characters" is now Your AIs - characters are one thing you can make with an AI, and the section covers creating, shaping, memory, and knowledge. Models and Routing are separate sections: routing gets its own spine - how it decides, the modes, health questions, attachments and images, and every setting in one place. Memory is its own page with the three kinds, what a memory is on the record, and the Memory page's tabs. Projects gain Permissions and approvals and The working session. Models documents "Best for this computer" and the maker labels on every card; Your records notes the model-file fingerprint each on-device answer carries. Old links redirect.

v1.2

Health questions: your model, your choice

The models guide now describes health routing as it ships in Your Own AI 0.5.0: health questions are answered on your device by default, you choose which downloaded model answers them, downloading a medical specialist offers the switch once, and choosing an online model means being asked before each question leaves your device.

v1.1

Recovery, as the app actually guides it

Backups and recovery now documents the guided flow that ships in Your Own AI 0.1.1: restore the Vault first, answer its import question before opening your apps, and follow the two-step key restore when the Vault holds a different conversation key. Also notes that held or failed backups announce themselves in Settings.

v1.0

The documentation site

The first release of docs.yourownai.net: getting started, characters (including memory, knowledge, and signed character packs), projects, models and routing, your data (records, receipts, backups and recovery), and the local endpoint for external apps.