Skip to content

Repository files navigation

Aliquot — from a scientific question to an experiment worth reviewing.

Open Aliquot · Run locally · Architecture · API & MCP examples

Give a scientific idea somewhere to go

Aliquot turns a plain-language hypothesis into a structured experiment plan. It searches the literature, drafts a protocol, looks up materials, checks catalog references, and presents the result in a desktop workspace you can review and revise.

The goal is to make the path from “what if?” to “how would we test it?” easier to inspect.

What a plan includes

Part What you can inspect
Literature check Related work, citations, and the model's assessment of novelty.
Protocol Proposed steps, parameters, and source references.
Materials & equipment Suggested reagents, supplier/catalog information, and verification results.
Budget & timeline Planning estimates in the selected currency and an execution schedule.
Validation & confidence Proposed checks and model-generated confidence annotations.
Scientist review Corrections to this experiment or reusable guidance for later plans.

Each experiment has its own draggable windows. A library keeps saved work accessible; attachments can add PDFs, images, and text to an experiment's context.

Three phases, visible progress

flowchart LR
    H["Hypothesis + attachments"] --> A["1 · Explore<br/>validate · classify · literature"]
    A --> B["2 · Draft<br/>materials search · plan synthesis"]
    B --> C["3 · Check<br/>catalog recheck · confidence"]
    C --> P["Reviewable plan"]
    P --> R["Scientist feedback"]
    R -->|"this experiment"| P
    R -->|"general guideline"| B
    classDef phase fill:#eee2d7,color:#463226,stroke:#b69276;
    class A,B,C phase;
Loading

Phase 1 streams events over SSE. The client then calls generation and verification separately, so a draft can be displayed while checks are pending. The current pipeline uses Claude, OpenAlex (plus arXiv for physics and climate), Tavily, and Supabase.

A “verified” catalog reference means an automated check found supporting catalog evidence. It does not certify experimental safety, supplier availability, or scientific validity. Confidence labels are model judgments, not calibrated probabilities.

Run locally

Use Node.js 20.9+, npm, and your own Anthropic, Tavily, and Supabase accounts.

git clone https://github.com/DocMorphic/aliquot.git
cd aliquot
npm ci

Create .env.local using the setup guide, apply supabase/schema.sql to your own Supabase project, then run:

npm run dev

Open localhost:3000. API providers may charge for generation and search; the repository does not include credentials or prepaid usage.

Feedback that carries the right scope

flowchart TB
    F["A scientist submits a correction"] --> S{"Which scope?"}
    S -->|"Experiment"| E["Revise this plan<br/>retain an audit record"]
    S -->|"General"| G["Save a domain guideline"]
    G --> N["Retrieve guidance for future plans"]
Loading

General guidance is retrieved by domain, rating, and recency. This is context reuse, not model-weight training. The schema includes vector columns, but the current guidance retrieval does not use embedding similarity.

Built for inspection

Layer Implementation
Desktop Next.js 16, React 19, Tailwind CSS 4.
AI orchestration lib/ai/ and app/api/experiment/.
Research Literature and catalog adapters in lib/search/.
Persistence Supabase Postgres and Storage through lib/supabase/.
Review Experiment corrections and reusable guidelines.

The architecture reference follows the implemented routes. The in-app Docs window includes HTTP examples and an MCP server stub; it is not a separately deployed MCP service.

Development

npm run lint
npm run build
npm start

The current application is a shared demo without per-user authentication. Its database schema permits public reads. Use non-sensitive demonstration data until access controls are implemented. Scientific plans require qualified review before laboratory use.

Further reading

No repository-wide license is currently declared. The cover is original SVG artwork illustrating a planning workflow.

Start with a question. Keep the evidence attached.

About

Turn scientific hypotheses into literature-backed experiment plans, with catalog checks, budgets, and scientist feedback in a desktop workspace.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages