Ada

The AI research agent for your terminal.

Ada is an AI-powered research agent that runs in your terminal. Ask questions, search papers, run experiments, write drafts, and produce cited research artifacts. Built on Pi with 197 scientific skills across 22 disciplines.

Install

git clone https://github.com/saikrishnabolla/ada.git
cd ada
npm install
npm run build
npm run dev

On first launch, Ada will install packages and walk you through model setup.

Requires Node.js 20.18.1 or newer.

Usage

Just ask anything:

$ ada "what do we know about scaling laws?"

Or use workflows for structured research:

$ ada deepresearch "mechanistic interpretability"
$ ada litreview "mRNA vaccine efficacy"
$ ada audit 2401.12345
$ ada replicate "chain-of-thought improves math"

Inside the REPL, type naturally or use /commands:

/deepresearch <topic>     thorough multi-source investigation
/litreview <topic>        literature review with citations
/review <file>            simulated peer review
/audit <paper>            verify paper claims against its codebase
/replicate <paper>        replicate a paper's experiments
/compare <topic>          compare sources on a topic
/draft <topic>            write a paper-style draft
/autoresearch <idea>      autonomous experiment loop
/help                     all commands

What Ada can do

Research -- Search papers via AlphaXiv and Semantic Scholar, traverse citation graphs, run literature reviews with verified citations, and produce source-grounded briefs.

Write -- Generate LaTeX papers, conference posters, slide decks (PPTX), Word documents, and publication-quality figures. Every claim is backed by a source.

Experiment -- Run code locally, in Docker containers, or on cloud GPUs (Modal, RunPod). Replicate paper results with isolated execution environments.

Analyze -- 197 scientific skills covering genomics, proteomics, drug discovery, clinical research, chemistry, physics, ML, finance, geospatial, and more. 325 pre-built workflow procedures across 22 disciplines.

Agents

Four specialized subagents, dispatched automatically for complex tasks:

| Agent | Role | |-------|------| | Researcher | Gather evidence across papers, web, repos, docs | | Reviewer | Simulated peer review with severity-graded feedback | | Writer | Structured drafts from research notes | | Verifier | Inline citations, source URL verification, dead link cleanup |

Skills

Ada ships 197 skills organized into:

• Research tools -- AlphaXiv paper search/Q&A, Semantic Scholar citation graphs, Crawl4AI web scraping, session search
• Scientific domains -- Genomics (scanpy, biopython), drug discovery (rdkit, diffdock), clinical (clinvar, pubmed), proteomics, chemistry, physics, materials science, neuroscience, ecology, astronomy
• Databases -- 250+ scientific databases including PubMed, ChEMBL, UniProt, COSMIC, ClinicalTrials.gov, OpenAlex, arXiv, KEGG, STRING, GnomAD
• Writing & visualization -- LaTeX papers, posters, PPTX slides, DOCX, PDF manipulation, matplotlib, seaborn, plotly, Mermaid diagrams
• Compute -- Docker sandbox, Modal (serverless GPU), RunPod (persistent GPU), Python REPL
• Workflow catalog -- 325 pre-built procedures from data analysis to quantum circuit design

Packages

27 Pi packages provide tools, agents, and integrations:

| Category | Packages | |----------|----------| | Research | AlphaXiv, arXiv, Tavily web search, session search | | Agents | Subagents, task tracking, ultrathink review loops | | Documents | Document parser, markdown preview, PowerPoint, Zotero | | Visualization | Charts, Mermaid diagrams, Excalidraw canvas | | Compute | Python REPL, sandbox, process management | | Infrastructure | MCP adapter, command palette, context management |

Configuration

On first launch, run setup inside Ada to configure services interactively. Or edit ~/.ada/.env directly:

# AI model (required -- at least one provider)
ANTHROPIC_API_KEY=
OPENAI_API_KEY=

# Web search
TAVILY_API_KEY=

# Scientific databases (optional)
S2_API_KEY=                    # Semantic Scholar
OPENALEX_API_KEY=              # OpenAlex
NCBI_API_KEY=                  # PubMed/NCBI

# Citation management
ZOTERO_API_KEY=
ZOTERO_LIBRARY_ID=
ZOTERO_LIBRARY_TYPE=user

# Cloud compute (optional)
MODAL_TOKEN_ID=
MODAL_TOKEN_SECRET=
RUNPOD_API_KEY=
OPENROUTER_API_KEY=

Global config at ~/.ada/.env works from any folder. Project-specific .env overrides per directory.

Architecture

ada/
├── packages/alpha-hub/     AlphaXiv CLI and library
├── extensions/             Research tools (/help, /init, /outputs)
├── prompts/                Workflow templates (11 orchestrated workflows)
├── skills/                 197 scientific skills (SKILL.md files)
├── .ada/
│   ├── agents/             4 subagent prompts
│   ├── themes/             Ada terminal theme
│   ├── settings.json       Package configuration
│   └── SYSTEM.md           Agent personality and rules
├── src/                    TypeScript source
└── bin/ada.js              CLI entry point

Built on the Pi agent runtime and Feynman by Companion AI. Scientific skills sourced from Claude Scientific Skills (MIT).

Development

git clone https://github.com/saikrishnabolla/ada.git
cd ada
npm install
npm run dev          # Run from source
npm run build        # Compile TypeScript
npm test             # Run tests (25 tests)
npm run typecheck    # Type checking

License

Apache 2.0 -- see LICENSE.

thinkwithada.com