Your next project starts with you

Stop repeating
yourself.
Make a playbook.

You’ve already told AI how you like things built. Gather those conversations on your own computer, then ask your agent to turn the patterns into instructions you can edit and reuse.

Give your agent a starting point See how it works ↓

Public source · Runs locally · Your own AI assistant

One set of standards. Two useful versions.

Your app playbook THE EVERYDAY VERSION

Build it the way
I like it.

Give mobile the same attention.
Make the main journey work.
Include polished social previews.

The same rules WITH EVIDENCE

Original messages. Projects. Dates. The context behind each preference.

↗ Open the source message

Illustration only. Your agent finds your own preferences.

↳ Codex + Claude local histories◎ All your profiles, one identity≡ Project and conversation timelines↗ Editable files you keep

From things you’ve said
to standards you can use.

The collector does the organizing. Your agent does the interpretation. You decide what belongs in your playbook.

Collect your words

The local collector reads supported Codex and Claude history files. It groups your human messages by project and conversation, across your profiles.

Keep the context

Browse each message with the preceding final AI reply. A short “implement that plan” makes more sense when the plan is right above it.

Make it yours

Have your own agent find repeated requirements, resolve changes in your preferences, and write two drafts. Edit them before using them on your next project.

No automatic playbook generator is built into the app. The prompt below gives your agent the complete local setup and drafting task.

The easiest way to begin

Let your agent
set it up.

Open a coding agent on the computer where you use Codex or Claude. Give it this prompt and let it handle the local setup.

Copy the prompt. Use an agent that can work with local files and run commands.
Let it collect your history. It should explain which folders it found and report anything missing.
Review your two playbooks. Keep the clean version for everyday use; open the evidence version when you want to check a rule.

Windows, macOS and Linux launchers are included. Python 3.11+ is required. No paid account with this project is needed.

Paste this into your local agent

Help me create my own app-development playbook from my local AI conversations.

Get the public source: https://github.com/slobodanmargetic988/WhoIamWithAI
Clone it to a normal folder on MY computer (or reuse an existing clean checkout). Read README.md, docs/AGENT_SETUP.md and docs/PLAYBOOK_FORMAT.md before working. This is a LOCAL archive, not a web app to deploy.

1. Set up Python 3.11+ with SQLite FTS5 if needed, using official software sources and the normal permissions of my agent. No administrator access or cloud accounts should be necessary for the app itself. On Windows use py -3; on macOS/Linux use python3. Start with the documented start.py launcher. Keep all existing conversations and any previous archive intact.

2. Discover and import MY supported local Codex and Claude history across my profiles and backups, treating all profiles as one person. Read histories only; do not open credentials or change original transcripts. Inspect discovery results and use explicit --source paths for missing supported folders. Run inside WSL for histories stored there. Clearly report inaccessible or unsupported sources, uncertain human authorship, partial imports and missing context. Never claim to have collected cloud-only chats or another OS user's history.

3. Open the local preview at http://127.0.0.1:8765 and verify projects, conversations, chronological human messages and available previous final AI replies. The server must stay bound to localhost. No public deployment, tunnel, uploads or analytics. Do not commit or publish my data, playbooks, exports, private paths or credentials. If you need a separate external AI service, ask me before sending any archive content to it; make clear how your own chosen assistant processes local files.

4. Analyze my human messages together with their preceding final AI replies. Find requirements I repeat across genuinely different projects, including short follow-ups that adopt a plan. Do not count duplicated profiles, forks or worktrees as independent evidence. Exclude uncertain/automated inputs. Read the full relevant messages and context; AI suggestions alone are not my requirements. Distinguish broadly repeated preferences, explicitly adopted global preferences and product-specific decisions. Resolve later changes in my preferences. If evidence is thin, say so instead of inventing rules. Do not include secrets or private contact details in either output.

5. Produce TWO editable drafts with the SAME rule IDs, requirements, scopes and completion checks:
- data/playbook.md: a clean, self-contained playbook WITHOUT evidence, quotes, citations, source links, message/thread IDs, source timestamps, research notes or personal project names. Make this the default version I can give an agent when starting a new app.
- data/playbook-with-evidence.md: the same rules PLUS supporting messages, project labels, timestamps, short verbatim quotes, relevant previous-AI-reply context and local links to the exact source messages. Use the documented thread-specific evidence links. Explain conflicts, qualifications and missing context here.
Also prepare data/customer-facing-app-playbook.json in the documented preview format. Run python3 -m ai_archive.playbook --write-outputs (Windows: py -3 -m ai_archive.playbook --write-outputs) to validate citations against the local archive and write both Markdown/TXT outputs. Fix validation failures rather than fabricating evidence.

6. Verify both files exist, their rule IDs and core instructions match, the clean version has no evidence, and evidence links resolve to the correct human messages and thread-specific AI context where available. Open /playbook in the local app and show me the clean download and the separate evidence download. Tell me where the editable files are and what I should review. Leave these as drafts for me to edit; do not silently install them as binding instructions in other projects.

Setup, collection, context review & both outputs

You can also select and copy the text directly.

Prefer to set it up yourself?

Open the public repository ↗ and clone it, or download the ZIP and extract it. Install Python 3.11 or later if it is missing. The app has no additional Python packages.

Windows

cd path\to\WhoIamWithAI
py -3 start.py --collect --open

Or double-click Start.bat and choose to collect your messages and open the app. The launcher also checks for python. Python must be installed first.

macOS / Linux

cd path/to/WhoIamWithAI
python3 start.py --collect --open

On macOS use Start.command; on Linux run sh Start.sh. The menu lets you collect again or open the existing archive.

Open http://127.0.0.1:8765 after the local app starts. Then give your agent the drafting instructions above. Keep the terminal open while browsing; press Ctrl+C there to stop the server. Imports can take time when your history is large.

Scope: collection works with supported local history files, not every cloud chat service. Native Windows execution has not been verified on this project’s current development machine. For Claude Desktop on Windows or histories inside WSL, ask your agent to check source locations and format support rather than assuming everything was found.

Use the rules.
Keep the receipts.

You don’t need to read the research every time. Both versions describe the same preferences, with different levels of detail.

Start here · No evidence

Your everyday playbook

Clear, scoped requirements and completion checks. No message quotes, source links or research notes. Edit it, then give it to an agent before your next build.

data/playbook.md

For reviewing the reasoning

The evidence version

The same rules, with the messages, dates and AI reply context that support them. Follow local links to inspect the original conversation whenever a rule needs a second look.

data/playbook-with-evidence.md

A guide on the web. An archive at home.

Your history doesn’t
come to this website.

This is a static instruction page. It has no account system, upload form, collector service or connection to your local archive.

The app runs on your computer.Collection and browsing need no remote service or AI API key. Your archive and generated playbooks are stored locally and excluded from Git.
You choose your AI assistant.A cloud-backed agent may process content through its provider when it reads your files. A local archive doesn’t automatically make that analysis offline.
The code is public. Your history is yours.Inspect or share the source repository. Keep your own database, transcripts, exports and playbooks off public repositories and web hosting.

A few useful things to know.

Will this generate my playbook as soon as I install it?

No. The app collects and previews your local history. Your own AI agent performs the analysis using the prompt above. It writes editable drafts and validates their evidence; you decide what to keep.

What if I don’t have enough repeated requirements yet?

Your agent should say so. It can separate a few supported preferences from ideas that need more evidence, or explain that the archive is too small to produce a useful playbook. A confident-looking list is not a substitute for your actual choices.

Can I collect new messages later?

Yes. Choose collection again in the launcher. Imports reuse the archive and skip unchanged source files. Ask your agent to revise both playbooks together when you want to include new conversations or changed preferences.

Do I upload this whole project to my web host?

No. This public guide needs only this index.html. The collector, database, local preview and playbooks stay on each user’s computer. Social preview artwork is served from the public source repository; no extra hosted app or server is needed.

Start your next app
with less repetition.

Give your agent the instructions. Make the first draft your own.

Get the setup prompt

Stop repeatingyourself.Make a playbook.

Build it the wayI like it.

From things you’ve saidto standards you can use.