big-tools
v0.1.0
Published
Between-sessions tools for the BIG methodology (Generate / Develop / Select / Promote). Developed for the Brown Practicum 'Ways to Discover a Good Idea'.
Maintainers
Readme
big-tools
Small tools for bigger ideas — between sessions.
What this is
Four phases. Nine tools that work today, a handful more on the way.
Every good idea goes through four phases — you generate it, you develop it, you select which ones to act on, and you promote the ones you kept. This is Carl Moore's BIG methodology, and it has been running groups to better ideas for forty years.
big-tools gives you small helpers for the work that happens between sessions — when the room is empty, the whiteboard has been erased, and the ideas need somewhere to live until the next time you meet.
| Phase | Tools | | ------------ | ------------------------------------------------------------------------- | | Foundation | Tracker | | Generate | NGT Round-Robin · Reframe · Mixtape · Steal · Ideawriting* · Selfish Explorer* | | Develop | Paper House · Delphi* | | Select | Filter · Discovery × Actionability* | | Promote | Pitch · Easy on People* · Between Sessions* |
* Planned. Running one today prints a friendly note explaining it's not built yet and suggests a working tool that's close.
Each tool is named after something a facilitator already does. The tool does not replace the facilitator. It just remembers what happened, and stands ready for the next time.
You run them on your own computer. Nothing you type goes anywhere except to the language model you choose.
Setup (technical section)
Everything below this line is for students who want to install the package and connect it to their preferred LLM client. If you are a teacher or facilitator planning to use big-tools in a class, you can stop reading here — the setup work is a one-time step your students will do themselves.
Install
npx big-tools@latest --helpThat prints every tool (including the stubbed ones, marked · stub) and shows
how to run each one.
Bring your own model
big-tools does not include a language model. You supply one through OpenRouter, which routes a single API key to dozens of models — including free-tier models that cost nothing for classroom use.
# 1. Sign up at https://openrouter.ai (free)
# 2. Create an API key
# 3. Tell big-tools about it:
big-tools config set openrouter-key sk-or-v1-...Other providers (OpenAI, Anthropic, Google, Ollama) are not supported in v0.1.
If an ANTHROPIC_API_KEY / OPENAI_API_KEY / GOOGLE_API_KEY is already in
your environment, big-tools will detect it and point you at OpenRouter
instead — OpenRouter can route to the same models with your existing key
routing, and it keeps the config surface small.
Pick a capability tier
Different tools want different models. mixtape wants a capable model;
ngt is fine on a free one. Instead of picking a model per call, you set a
capability tier and big-tools picks the model for you:
big-tools config set tier free # default — all free-tier models
big-tools config set tier moderate # mid-range models (small per-call cost)
big-tools config set tier full # frontier models (higher quality, higher cost)When a tool's preferred tier is higher than yours, it runs on your tier anyway and prints a one-line note telling you what you'd gain from an upgrade. No tool ever refuses to run because of tier.
Set a budget (optional but recommended)
big-tools config set budget 10.00 # monthly cap in USD
big-tools config set budget unlimited
big-tools config get usage # spending summary for the current monthBudget accounting is estimated locally from OpenRouter's published pricing. Once your monthly ledger crosses the cap, further calls refuse until you raise the budget. The ledger rolls over on the 1st of each month.
Manual model override (power users)
big-tools mixtape "my idea" --model anthropic/claude-3.5-sonnetAn explicit --model skips tier selection and fallback. If the model is
unavailable, the call fails instead of silently switching.
Run a tool
big-tools tracker add "students skip the warmup when they don't name the person they're serving"
big-tools reframe "students skip the warmup"
big-tools mixtape "the shape of between-sessions work"
big-tools filter purpose --name default
big-tools filter "accessible EEG visualization"
big-tools pitch 1 --to funder
big-tools --helpEvery tool accepts --json for scripting and --model <id> for one-off model
overrides.
Privacy
Read PRIVACY.md before you use the package in a class. Short
version: big-tools runs entirely on your computer. The only thing that leaves
your machine is the prompt text you send to OpenRouter (and from there, to
whichever model OpenRouter routes to). big-tools keeps a local usage ledger at
~/.config/big-tools/usage.json with tool names, model IDs, and token counts
— never the prompts themselves.
License
Code is MIT. The corpus in corpus/ is under a separate license — see
corpus/LICENSE for details.
big-tools is a project of the Brown Practicum "Ways to Discover a Good Idea" (Spring 2026).
