Becki: One memory. Every AI you use.

01The problem

Your context is scattered across a dozen models that don't talk.

Every tool starts from zero. You re-paste your project, your preferences, your past. The smarter your tools get, the more amnesia you carry.

I.

Every chat starts blank.

You explain your codebase to Cursor, then Claude, then Codex. Again tomorrow. The ones with memory don't share it. Claude Code can't see what you told Claude Desktop, same company.
II.

Preferences don't follow you.

The voice you taught one model, the constraints it learned, the rules of your taste. Invisible to the next one you open.
III.

Knowledge dies in tabs.

Insights live in a chat window you'll close in twenty minutes. Decisions live in a meeting nobody re-watches. Your future self never sees them.
IV.

You are the glue.

The only thing connecting your tools is you. You carry context from one window to the next, by hand, every day. The work that should compound just evaporates.

02How it works

Record once. Remember everywhere.

Becki captures your meetings, distills what matters, and serves it to every AI tool you use over MCP. A small daemon on your Mac. Three jobs, done quietly.

Meetings

Calendar

Notes

↓ capture ↓

becki.local · transcribe + index

Step 01 · Capture

Record every meeting.

Becki auto-starts when a calendar meeting begins, or push the menu-bar button. On-device transcription, speaker diarisation, project tagging. No audio ever leaves your Mac.

"ok so we should probably push the release back a week, Lina you'll send the onboarding doc by Friday right, and we still need to decide on the MCP terms thing, I'll ask legal, also let's drop xterm and just use the…"

→

release: push 1 week

Lina → onboarding doc Fri

me → ask legal on MCP terms

decided: drop xterm

project: studio-shell

Step 02 · Distill

Commitments, not transcripts.

Becki extracts the durable bits: who owes what, what was decided, which project it belongs to. The transcript stays for reference. The graph stays small and meaningful.

› claude · what did we decide about xterm?

becki MCP · 12 ms + decided 3d ago: drop xterm
+ context: studio-shell, eng sync
+ follow-up: 68b290f, open

You decided to drop xterm during the Eng sync three days ago, in favour of the lighter shell. There's still an open follow-up to confirm the v5 upgrade path.

Step 03 · Serve

Every AI, in context.

Becki exposes an MCP server. Claude Code, Cursor, Codex, Zed, VS Code, and anything else that speaks MCP can ask "what did we decide?" and pull the answer from your own memory.

03Works with

Plugs into anything that speaks MCP.

Becki exposes a local MCP server. Your AI tools connect to it the same way they'd connect to any other. No extension, no proxy, no Becki account.

Becki lives in

Claude · desktop

Claude Code · CLI

Cursor · editor

Codex · CLI

Gemini · CLI

Zed · editor

VS Code · editor

Raycast · launcher

ChatGPT · desktop

Claude.ai · web

Perplexity · web

+Any MCP client · bring your own

Becki knows where it lives

GitHub · repos & PRs

Gmail · mail

Slack · threads

Linear · issues

Notion · docs

Your AI tools already reach into GitHub, Gmail, Slack, Linear, and Notion through their own MCP servers. Becki captures the map: what got decided, where the spec lives, who owns the doc. So the next session opens with it already in hand.

04Local-first

Your memory lives on your machine.

Becki runs on your Mac. Your meetings transcribe on-device and your vault stays in files on your disk. No opaque "memory feature" controlled by someone else's product team.

Meetings transcribe on-device. Raw audio and your voiceprints never leave your Mac.
Your vault lives in files on your disk, encrypted at rest.
Apple silicon native. A small, quiet daemon, not a background hog.
Encrypted, opt-in sync across your own devices.
No screen capture. No always-on watcher. No data sold.

Your Mac

becki.local

on your machine

claude

codex

cursor

— local-first —

Becki HQ your memory isn't our product.

05 How Becki compares

Other tools remember in one place. Becki remembers everywhere.

Meeting recorders capture the room and forget it. Note apps only hold what you stop to write down. Every AI keeps its own memory and shares none of it. Becki is the layer underneath, on your Mac and in every tool.

	Per-tool memoryChatGPT · Claude Projects	Meeting recordersGranola · Otter · Fireflies	Note appsNotion · Mem · Reflect	Becki
Shared across every AI tool	Locked to one	—	—	Yes, via MCP
Runs locally on your Mac	Cloud	Cloud	Cloud	Yes
Captures decisions & dead endsnot just transcripts or notes you write	Partial	Raw transcript	You write it	Automatic
Surfaces context inside your toolsno copy-paste between apps	Within itself	—	—	Everywhere
Sharpens as it growsconsolidates instead of piling up	—	—	—	Yes
The memory is yoursportable, not vendor-locked	Vendor-locked	Vendor-locked	Exportable	Yours, on disk

Your context is scattered across a dozen models that don't talk.

Every chat starts blank.

Preferences don't follow you.

Knowledge dies in tabs.

You are the glue.