VCN #48: TOTAL RECALL - BEFORE DOORS
Repo memory, retrieval and context graphs for your coding agent
Saturday 2026-08-08, 10:00-13:00 PT, Frontier Tower Floor 9, San Francisco

This file is agent-readable on purpose. Point your coding agent at it and let it
do the setup, or run the commands yourself. Ten minutes, done before you arrive.


WHAT YOU NEED IN THE ROOM
-------------------------
1. A laptop. Laptops are mandatory for this one - it is a build sprint, not a talk.
2. A repo you actually work in. Bigger and messier is BETTER: you already know
   which questions are hard to answer in it, and those are the questions the
   morning is built around. A Python repo is the smoothest path (step 1 uses
   Python's built-in parser); any language works, see NOT PYTHON below.
3. Python 3.10+ and `uv`. If you do not have uv:
       macOS / Linux : curl -LsSf https://astral.sh/uv/install.sh | sh
       Windows       : powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
4. Your ticket. It carries the two things that make the labs free for you:
       - z.ai + Claude Code access for the session
       - a Nebius Token Factory key, which runs every embedding in step 2


TEN-MINUTE PREFLIGHT (do this at home)
--------------------------------------
From inside the repo you are bringing:

    mkdir -p .recall && cd .recall
    uv init -q && uv add -q openai chromadb

Then confirm it works:

    uv run python -c "import chromadb, openai; print('ok')"

That must print:  ok

If it does not, bring the error to the door at 10:00 and we will fix it before
the sprint starts. This one line is the hard gate; nothing in the morning works
until it passes, and the worst outcome is discovering it at 10:50.


WHAT YOU DO NOT NEED TO DO YET
------------------------------
Do not set the API key at home. You get it on your ticket in the room:

    export NEBIUS_API_KEY=...          # PowerShell: $env:NEBIUS_API_KEY = "..."
    export EMBED_BASE_URL=https://api.studio.nebius.com/v1

Do not pre-index anything. Watching your own repo get chunked is the point of
step 1, and a pre-built index skips the part where you look at the chunks and
find out they are wrong.


NOT PYTHON?
-----------
Step 1 splits code on real function and class boundaries. It uses Python's
built-in `ast` module because that needs no build toolchain and cannot strand
you on install. For any other language the shape is identical with tree-sitter,
which covers JavaScript, TypeScript, Go, Rust and about forty more with the same
node types. If you want to get ahead:

    uv add -q tree-sitter tree-sitter-language-pack

Bring the repo you care about either way. Swapping the parser is one function.


WHAT YOU LEAVE WITH
-------------------
A `.recall/` folder on a repo you actually work in, containing:
    chunk.py     splits the codebase on function boundaries, keeping path + symbol + line
    index.py     embeds those chunks into a vector store that persists on disk
    ask.py       returns the right files by meaning, formatted for the agent to cite
    edges.json   answers "who calls this", which no vector index can
    MEMORY.md    notes the agent reads at startup, so what it learns survives

The proof is a brand new session: close everything, open a fresh one, and ask
the question your agent has never once got right.


RUNNING ORDER
-------------
    10:00   doors
    10:15   walkthrough (the four kinds of memory, where vectors stop working)
    10:45   build sprint, steps 0 to 4
    12:00   demos and coffee
    13:00   out


Tickets: $10 early / $20 door. Frontier Tower members free, reach out to the
team directly for a ticket.
RSVP: https://luma.com/vcn-48-total-recall
Telegram: https://t.me/+EBFzKXmJAVk5ZGU0
Site: https://vibecodingnights.com

Hosted by Vibe Coding Nights: Rayyan Zahid (Immersive Commons), Michalis
Vasileiadis (Hacker Bob), Eric Mockler (AI Geneticist), Devinder Sodhi
(Learning Layer Labs). Facilitator: Rayyan Zahid.