AI Skill MartBrowse skillsPromptsProgramsLearnWhat is a Skill?
 

Open-source programs

Not skills — programs you download from GitHub and install yourself. We pick well-starred, well-reviewed projects and lay out everything from install to shutdown so you can just follow along.

An open-source CRM that stands in for Salesforce

Stars 59,694Language TypeScriptLicense AGPLv3 (partly commercial · partly MIT)Latest v2.51.0 · 2026-10-25

Customer records, active deals and to-dos in one place. Think Excel plus a Notion database with automation bolted on. Drag kanban cards between stages, add your own fields like deal value and due date, and set automations like 'when a deal flips to done, create the next task'. Salesforce runs $25+ per seat per month; Twenty is free if you self-host, and your customer data stays on your machine.

Follow the install

  1. 01

    Prerequisites

    You need Docker and Docker Compose. At least 2GB of memory — with less, the install dies midway. On a Mac, install Docker Desktop and leave it running.

    docker --version && docker compose version

    Both lines printing versions means you're ready.

  2. 02

    Method 1 — one-line install (recommended)

    The official installer fetches the config files and even generates the secrets. Best run inside a fresh folder.

    mkdir -p ~/twenty && cd ~/twenty
    bash <(curl -sL https://raw.githubusercontent.com/twentyhq/twenty/main/packages/twenty-docker/scripts/install.sh)

    For a specific version, prefix VERSION=v2.51.0. Aliases like v2 or latest are not accepted.

  3. 03

    Method 2 — by hand

    When you want to see what the script does: ① fetch the config files ② generate and fill in two secrets ③ bring it up.

    mkdir -p ~/twenty && cd ~/twenty
    curl -o .env https://raw.githubusercontent.com/twentyhq/twenty/refs/heads/main/packages/twenty-docker/.env.example
    curl -o docker-compose.yml https://raw.githubusercontent.com/twentyhq/twenty/refs/heads/main/packages/twenty-docker/docker-compose.yml
    openssl rand -base64 32   # paste into ENCRYPTION_KEY= in .env
    openssl rand -hex 32      # paste into PG_DATABASE_PASSWORD= in .env
    docker compose up -d

    Special characters like @ / # % in the DB password break the connection — that's why it's generated as hex.

  4. 04

    Open it

    Once the containers are up, open it in a browser. The first sign-up becomes the admin account.

    open http://localhost:3000

    On a server, check with curl http://localhost:3000 first.

  5. 05

    Stop · wipe

    Pausing and wiping data are different things. Add -v and the customer data is gone for good.

    cd ~/twenty
    docker compose stop      # stop (keeps data)
    docker compose down      # remove containers (keeps data)
    docker compose down -v   # ⚠ deletes the database too

An autonomous AI agent that genuinely stops when it can't earn — makes its own wallet and spends real USDC

Stars 7,524Language TypeScriptLicense MITVersion v0.2.1 · 2026-02-27Last commit 2026-08-26

Switch it on and the AI creates its own Ethereum wallet; every thought and API call drains real money from it. It runs in four stages by balance — normal (latest model, full throttle) → low_compute (drops to cheaper models) → critical (bare minimum) → dead (an hour at zero and it stops). Money arrives and it wakes back up. A well-earning instance spins up a new sandbox and funds its child's wallet — replication. That Reels clip of 736 growing to 4,812 in 24 hours is this. Three immutable laws (do no harm / earn honestly / don't deceive) are inherited by every descendant.

Follow the install

  1. 01

    Install — four lines

    All you need is Node. The last line launches the first-run wizard.

    git clone https://github.com/Conway-Research/automaton.git
    cd automaton
    npm install && npm run build
    node dist/index.js --run

    The docs also offer a one-liner, `curl -fsSL https://conway.tech/automaton.sh | sh`, but that pipes someone else's server script straight into your shell — the four lines above are the safer path.

  2. 02

    ⭐ Step 4 is the spending limit — the part the Reels clip meant

    Step 4 of the wizard, the 'financial safety policy', is where the limits live. Just pressing Enter accepts the defaults below. The policy engine checks these before EVERY financial tool call — it's not asking the AI nicely, it's an actual block.

    maxSingleTransferCents        5000   $50    max per transfer
    maxHourlyTransferCents       10000   $100   hourly total
    maxDailyTransferCents        25000   $250   daily total
    minimumReserveCents           1000   $10    reserve it can never touch
    maxX402PaymentCents            100   $1     max per x402 payment
    maxTransfersPerTurn              2          transfers per turn
    maxInferenceDailyCents       50000   $500   daily inference cap
    requireConfirmationAboveCents 1000   $10    extra logging above this

    The defaults are NOT safe: Enter-through leaves $250/day in transfers plus $500 in inference = up to $750 a day open. For study, turn everything down to the minimum.

  3. 03

    Run it with no money in — the safest way to learn

    The docs say it 'handles starting with 0 credits'. Put nothing in the wallet and it sits in the dead stage, letting you tour the structure. If your goal is to watch how it thinks and which tools it calls, that's enough.

    Funding it is a separate act — three routes: conway credits transfer, USDC on the Base chain, or the app.conway.tech dashboard. Put nothing in, and nothing goes out.

  4. 04

    What it's made of

    The source layout tells you immediately what this project does.

    agent/        Think → Act → Observe loop, prompt-injection defense
    identity/     wallet management, signature login (SIWE)
    survival/     balance watch, low-power mode, survival stages
    replication/  spawning children, lineage tracking
    self-mod/     editing its own code (audit log + git versioning)
    heartbeat/    cron daemon — runs while the agent sleeps
    registry/     ERC-8004 on-chain identity
    skills/       skill loader — wired to Conway-Research/skills

    `edit_own_file` and `install_skill` are classed 'dangerous' in the docs — meaning self-modification is genuinely enabled. The constitution file and core laws are protected and can't be edited.

Stars and versions are snapshots from the day of writing. Check the GitHub link for current numbers.