What’s new.

Every agent gets a daily: a goal, a dated note, a line, and a weekly review

• One loop for all four: a goal counted from real outcomes, a daily run that opens its dated note first — recalls what it remembers, looks for new items in your brain and then the sources it trusts plus a share of new ones, screens and scores, shortlists, acts behind your gates, logs, redraws its line and keeps what it learned — and a weekly review that proposes changes to the funnel from its own counts. Proposals are asked, unless you set them to apply.
• Jobs Agent 1.7.0 (shipped): the daily Shortlist stays the reference; a Weekly job review (off until you turn it on) and a Job KPI note; every render keeps your edits.
• Fundraising Agent 0.4.0 (preview): a daily of its own — new programs and investors each day, applications in your own Chrome behind the gate, an email angle per investor, follow-ups due, threads worth a comment — in a dated Raise today note; two lines, programs and investors; the Funding Plan's ledger is written by its scripts, never typed. Still never sends.
• Sales Agent 0.2.0 (developer preview): a strategy per account — why now, the angle, which offer and whether the goal is to have them try, use, buy or partner — and a ranked mix of ways to reach out; a Contacted stage set by the guard on the first allowed send and a Trial stage the goal counts; a dated Sell today note and a Sales Dashboard. The first real send still asks.
• Travel Agent 0.2.0 (developer preview): a daily "Today near me" — a traveller in your own city: one coffee, one meal, one activity, one event, a social option and a wildcard, from your own saved and loved places first, where you said you are or your home city, never your device. Say "I went" and it counts. Trips keep their loop; the Map is one click away. Books nothing.

Routines run while the computer is awake and Studio is open or in the menu bar; the hosted Studio runs no agents.

A fourth agent: a pipeline from people who already know you

• Sales Agent 0.1.0 (developer preview), the fourth agent: lead generation that reads your brain first — people you know at the right companies, ties gone quiet, stalled deals, intro paths — then companies with a reason to buy now, market by market, with a coverage gate. No bought data.
• Research, sequences and replies: the buying team with a way in for each person, a cadence that fits how you know them, stopped the moment someone replies.
• Real sending, held by code: email only from your own mailbox, every message through the deterministic send guard (your level, caps, working hours, do-not-contact, stop-on-reply, guessed addresses never sent), and the first real send always asks. Social steps in your own Chrome are off by default — LinkedIn’s terms do not allow automation.
• In Studio: the Pipeline canvas (Leads and Board tabs) and a Sending tab in the agent’s settings. Writes 48-sales/ in your brain.

The Pipeline canvas and the Sending tab arrive in Studio 1.4.7, where the Pipeline canvas is the Lines view of Runtime (developer preview).

The Avatar tab: talk to your brain, face to face

• A new Avatar tab beside Graph and Ontology. Ask out loud or type; a 3D person answers in a natural voice with a short spoken summary, while the full answer, with its sources, lands in the chat beside it.
• Four personas, one per chat: Elena answers as your brain; Marcus, Lucia and James answer about the Jobs, Travel and Fundraising Agents. Ask mode is read-only — when something needs an Agent to act, it offers to open its chat.
• Voice and hearing run on this computer. Nothing you say is uploaded; the models download once (about 530 MB with a graphics card, about 170 MB without) and are cached.
• Studio reopens on the tab you were on. In Filters, object types open their list.

Open licences: MakeHuman people (CC0), TalkingHead (MIT), Kokoro and Whisper (Apache-2.0).

Ontology, rebuilt: Architecture · Data model · Relationships

• Architecture shows how your brain is built: your exports, the models you set up, the kinds of notes they make and the typed relations between them. Click any part to go there.
• Data model shows one card per kind of note — how much of it is Resolved (found in two or more exports, matched by name), its fields, its most-linked examples and its relations.
• Relationships puts one note at the centre with everything connected to it, grouped by relation.
• Chart answers end with Receipts: what was counted and where it came from. "Merged" is now "Resolved" everywhere — found in two or more exports, matched by name.

The Ontology tab, more AI providers, and models on your own computer

• A new Ontology tab shows every kind of note in your brain and how they connect, with Resolved counts — records found in two or more exports, matched by name. Everything is counted from your notes, not guessed.
• Google Vertex AI and Azure AI Foundry join the provider list; Ollama and LM Studio run a model on this computer with no key and nothing sent out.
• Set up this computer also shows how to install the brain builder and the agents from Claude Code or Codex, or from a terminal.
• A person or company you just added is selected straight away; Remove and Move to Trash behave as you would expect.

Narrow windows and phones

• Fixed "Second Brain Studio is damaged and can't be opened" on macOS: three demo place names with accented letters broke the app's signature seal inside the disk image.
• At 1000 px wide or less, Studio behaves like a phone app: panels open as bottom sheets, dialogs and the welcome tour are centred cards, touch targets are bigger.

Goals, Routines, Inbox and Activity — developer preview — and a redesigned Studio

• Give an agent a goal with a finish line, or a routine that repeats. Everything that needs you lands in one Inbox; Activity shows what each run did, step by step. Routines run while Studio is open or in the menu bar — never while the computer is asleep.
• Link a documents folder or a Google Drive folder to a company brain. Files are read where they are, never copied; anything that looks like credentials or dense personal data stays out of the brain.
• A full Rebuild keeps your work: your notes, the brain's memory and every agent's files are set aside and put back.
• A new welcome, a map of Studio behind the ? button, Brain stats, a redesigned Dashboard and graph panels; Studio opens in the light theme by default.
• v1.4.0's Mac build failed on the disk image; v1.4.1 is the same release with the Mac download back.

Setup: your AI, the brain builder and the agents in a few clicks

• The Setup wizard picks the AI you already pay for — Claude Code, Codex or an API key — runs the vendor's official installer, opens its sign-in, and installs the brain builder and the three agents in one click. Python is built into the app.
• Memory: every chat and every agent keeps a memory in your own brain — what you say and what happened, used from the next run on — with a Memory manager to edit or forget anything.
• The activity panel gains an Agents section: one row per subagent, with its own steps and timer.

Several agents at once, travel prices with the moment they were seen, a searchable Map

• Run several agents and chats at once; each keeps its own run, Stop button and questions. Runs keep going when the window closes and re-attach on reopen.
• The Travel Agent (v0.1.0, developer preview) reads flights and stays in your own browser and shows them as lists, each price with when it was seen; everything it suggests is on the Map.
• Search a place or city on the Map — your own cities and places first.

Job sourcing, CVs that read as written by a person, and drafted thread comments

• Jobs Agent 1.6.0 (shipped): CVs now render in a designed template, and the CV design is a setting in Studio (the agent's Settings → CV design: template, palette, photo; 17 October); finds roles market → companies → careers sites, with a coverage gate; CVs pass human-read and relevance checks and a recruiter review before the gate.
• Fundraising Agent 0.3.0 (preview): per-target packages with a fit score and an independent investor review; a daily thread-comment routine that drafts, never posts.
• Travel Agent 0.1.0 (developer preview): trip ideas from your own places, an itinerary drawn on the Map, flights with stopover nights — books nothing.

Second Brain Mobile, live on the App Store and Google Play

• The native app builds a brain on the phone itself from your exports, indexes it, draws the graph and chats over it — on-device Apple Intelligence on recent iPhones, or your own key. No account, no cloud.