Start building with local models
Just run:
npm startThat starts the chat REPL. First-time setup (Ollama, clone, dependencies, .env) is below if you have not done it yet.
Install Ollama (paste in terminal, or download Ollama):
curl -fsSL https://ollama.com/install.sh | shPull the default model, then this app:
ollama pull gemma4
git clone https://github.com/1dolinski/hammock-ai.git
cd hammock-ai
npm install
cp .env.example .env
# set PRIVATE_KEY (USDC on Base for APINow) — optional: OLLAMA_MODELThen npm start (see After install).
QMD (@tobilu/qmd) is installed by npm install and runs automatically — no global install needed.
Why QMD + Gemma 4 + APINow (x402) — Gemma 4 runs locally via Ollama. QMD hybrid-searches your markdown (memories, chats). APINow pays per API call with USDC / x402—one wallet, no per-vendor API keys.
Use: npm start · /tasks /memory /qmd /clear · quit · npm run start:verbose
LLM Wiki: AGENTS.md is the schema; committed wiki/ (compiled notes) and raw/ (immutable sources) are QMD collections llm-wiki and raw-sources. Tools: wiki_read, wiki_write, wiki_append_log, raw_read. Idea: Karpathy gist.
More: ABOUT.md
All advanced features are opt-in. Base npm start works exactly as before.
Schedule recurring prompts from chat:
"every morning at 8am, get my horoscope"
The model calls add_cron("0 8 * * *", "get my horoscope"). Jobs persist in SQLite (data/app.sqlite) and resume on restart.
Tools the model can use: add_cron list_crons remove_cron toggle_cron
npm install sets up data/app.sqlite automatically (via better-sqlite3, already bundled). The model can run read-only SQL with query_db:
"show me all my cron jobs that ran today"
Run the assistant as a Telegram bot instead of CLI:
- Talk to @BotFather and create a bot
- Add to
.env:
TELEGRAM_BOT_TOKEN=123456:ABC-DEF...npm start-- detects the token and starts polling Telegram instead of the REPL
Same tools, memory, and cron -- just a different frontend.
