PulSeed

An AI agent system that grows your goals from seed to tree.

Set a goal. Seedy takes it from there — observing, planning, delegating to AI agents, and tracking progress until it's done.

• Track long-term goals with real measurable progress — "Increase test coverage to 90%," "Double monthly revenue," "Keep my dog healthy." Seedy decomposes goals into dimensions, measures each one, and keeps going.
• Delegate work to any AI agent — Claude Code, OpenAI Codex, Browser Use, or your own custom adapter. Swap agents without changing goals.
• Know when to stop — Seedy applies satisficing: when every dimension crosses its threshold with enough evidence, the goal is complete. No infinite loops. No premature exits.
• Stay safe — Irreversible actions always need your approval. Trust is asymmetric: one failure costs more than three successes earn.

1. Install PulSeed (Node.js 20+):

npm install -g pulseed

2. Set your API key:

export OPENAI_API_KEY=sk-...

# Or use Anthropic
# export PULSEED_LLM_PROVIDER=anthropic
# export ANTHROPIC_API_KEY=sk-ant-...

3. Plant a goal and let Seedy grow it:

pulseed goal add "Increase test coverage to 90%"
pulseed run
pulseed status

Seedy assesses feasibility, breaks the goal down, delegates tasks to agents, and tracks progress automatically.

Using OpenClaw? Install the official plugin for seamless integration — see @pulseed/openclaw-plugin.

Code Quality — "Increase test coverage to 90%." Seedy observes current coverage, identifies untested modules, delegates test writing to a coding agent, and verifies with real test runs.

Revenue Target — "Double monthly revenue within 6 months." Seedy tracks revenue metrics, identifies growth opportunities, delegates research and implementation, and measures actual outcomes.

Health Monitoring — "Keep my dog healthy." Seedy monitors health indicators, schedules vet checkups, tracks nutrition, and escalates when human judgment is needed.

More examples in docs/usecase.md.

• Goal-driven orchestration — Set a destination, not step-by-step instructions
• Agent-agnostic — Swap agents without changing goals
• Satisficing — Knows when "good enough" is enough
• Asymmetric trust — Failure costs more than success; safety by default
• Stall detection — Detects loops and changes strategy automatically
• Plugin system — Extend with custom adapters, notifiers, and data sources
• Goal trees — Decompose large goals into sub-goals, each tracked independently
• TUI dashboard — Real-time terminal UI with progress, logs, and approval flow

Deep dive: Architecture Map | How it works

Custom adapters: Plugin Development Guide

Programmatic usage:

import { CoreLoop, StateManager } from "pulseed";

const stateManager = new StateManager("~/.pulseed");
const loop = new CoreLoop({ stateManager, /* ...adapters */ });
await loop.runOnce();

CLI reference: See docs/getting-started.md for the full command list.

Development setup:

git clone https://github.com/my-name-is-yu/PulSeed.git
cd PulSeed
npm install
npm run build
npm test

State: ~/.pulseed/ | Reports: ~/.pulseed/reports/ | Ethics logs: ~/.pulseed/ethics/

See CONTRIBUTING.md for guidelines.

See CHANGELOG.md for version history.

PulSeed stores all state locally. No telemetry. No phone-home. Your LLM provider is the only external connection.

MIT License

Plant the seed. Watch it grow.