Structured AI research

The reasoning ledger for serious research

PACT is a research tool that makes AI reasoning permanent, addressable, and reproducible. Not a chat. A structured notebook where every prompt and response becomes a citable, chainable artifact.

Try the Research Service Read the articles
See it in action

Real research. Real questions. Real PDFs.

Browse the samples collection to see what PACT research looks like — from the opening question through the guided conversation to the delivered document. Every sample is free to read and download.

Physics Medicine Music History Economics Cycling Research Methodology
Browse research samples →

Chat disappears. Research shouldn't.

Every AI chat response is ephemeral — when the conversation ends, the reasoning is gone. PACT treats each response as a permanent, addressable cell stored in a local database.

Chain cells across discussions using cell references. Each step receives the full prior response as context — not a summary. Build analyses that would be impossible in a single prompt.

[Cell 001] Medication list — purpose of each drugClaude · 4.2s · discussion-1
[Cell 002] Pairwise interactions ← [Cell 001]Claude · 8.4s · discussion-2
[Cell 003] Triple combinations ← [Cell 002]Claude · 12.1s · discussion-3
[Cell 004] All four together ← [Cell 003]Claude · 19.8s · discussion-4
[Cell 005] Clinical summary ← [Cell 004]Claude · 30.5s · discussion-5 · PDF →

Everything chat can't.

Built as a VSCode extension with local SQLite storage — your research stays on your machine.

Permanent cells

Every prompt and response is saved permanently. Restart PACT and your research is exactly where you left it.

Cell references

Reference any prior response in any new prompt — across discussions, across notebooks. Full context, not summaries.

System prompts

Each notebook has an optional system prompt. Set the AI's role once — it applies to every discussion silently.

Model tiers

Switch between Economy and Standard tiers globally. Use Economy for initial exploration, Standard for precision work — without changing your workflow.

Diff view

Compare two responses side by side with sentence-level highlighting. See exactly how models and tiers differ.

Notebook sharing

Export as signed .pact files. Other researchers import and continue exactly where you left off — with cryptographic verification.

Four articles documenting PACT.

From the first prototype to real medical research using progressive chaining.

01

What if AI conversations were structured like notebooks?

Introducing PACT — why treating AI responses as permanent, addressable cells changes everything about research.

02

Building PACT: a structured AI research tool

The architecture behind notebooks, discussions, cells, and the SQLite reasoning ledger.

03

How I used PACT to analyze my medication interactions — and why chat couldn't do it

Five progressive prompts, cross-discussion cell references, and an 84-page clinical analysis exportable for a physician visit.

04

Why Research Topology Matters: Star vs Chain in AI-Assisted Notebooks

Two alternative approaches to structuring AI-assisted research workflows.

Research notes.

Observations and updates from building and using PACT — less formal than a Medium article, more considered than a tweet.

Development

PACT v0.0.3 — what's new in the global menu

Model tiers, signed notebook export, config management, and the full story of what each menu item does.

Development

pactresearch.net is live

The PACT Research website launched today. A brief note on what it represents and what comes next for the research network.

Join the research network

PACT is in active development. Beta access is available for researchers, physicians, and domain experts who want to shape how structured AI research works.

Request access

Or email directly: nikolaj.ivancic@gmail.com