jumpyGoatHq - Control Center for Agents

Get Minimal AI Agents running automations for your business

jumpyGoatHq is for

• teams who want AI to work while they sleep
• what the whole team to have access to the same agents
• want to have several long running agents with different skills/context

Examples

• a marketing agent scheduling social media content every Monday
• a coding agent running a security check every night
• a bookeeper agent filing invoices as they come in

Each of those agents must know different stuff and have access to different tools and must do its work in different intervalls.

The only job jumpyGoatHq has is build the base for you to set that up easily and without knowing how to code.

The tag line is about minimal AI Agents. Why? Well, because most tools out there are way too complex. Bring too many features. Become a black box where you don't understand what agent is doing what.

How it works

What jumpyGoatHq does under the hood (technically)

A tiny, file-native control plane for Pi-powered agents.

Define agents in markdown, give them scheduled automations or one-off tasks, run them through Pi, and keep an auditable local history of what happened.

agents as markdown → schedules/tasks/operator commands → Pi runs → SQLite run history

working on adding smart memory system.

Screenshots

Reusable agents	File-backed task board

Core primitives

Primitive	What it means	Stored as
Agent	Reusable Pi runtime persona/instructions/context/policy bundle	workspace/agents/<agent>/AGENT.md
Automation	Manual or scheduled prompt for one agent	workspace/automations/<automation>.md
Board/task	Markdown kanban and assignable one-off work	workspace/boards/<board>/...
Run	Auditable execution receipt	workspace/data/jumpygoat-hq.sqlite
Connector/tool	Gated external capability such as web search or email	connector config + env secrets

Quick start

pnpm install
pnpm build
pnpm run doctor

Install and authenticate Pi as the same Unix user that will run jumpyGoatHq or cron:

npm install -g @earendil-works/pi-coding-agent
pi /login
pi --mode json --no-session "hello"

Optional local environment overrides and secrets go in .env.local:

cp .env.example .env.local
# edit .env.local if needed

This repo ships as a template with no active agents, automations, tasks, or SQLite data. By default mutable instance state lives under local workspace/ and is gitignored. Set JUMPYGOATHQ_HOME=/path/to/jumpygoat-hq-home to place that state somewhere else for deployment. For a VPS/systemd install, see docs/DEPLOY.md; for existing server updates/rebuilds, see docs/UPDATE.md.

Create your first agent and automation

Create workspace/agents/daily-review/AGENT.md:

---
name: daily-review
description: Reviews current work and produces a concise daily brief.
model: fast
allowedIntents: []
---

You are the daily review agent. Be concise. Identify blockers, due items, and recommended next actions.

Create workspace/automations/daily-review.md:

---
agent: daily-review
schedule: "manual"
---

Review the workspace notes and open tasks. Tell me what needs attention today.

Run it:

pnpm runner daily-review

Open the web UI:

pnpm web
# http://127.0.0.1:3000

The UI is intentionally raw server-rendered HTML. It shows agents, automations, boards/tasks, a kanban board, scheduled run agenda, cron evidence, runs, run details, settings, and a simple “Run now” button.

Scheduled automations

Automation schedules are either manual or 5-field cron expressions in automation frontmatter:

schedule: "0 9 * * *"

Install, inspect, or remove local cron entries:

pnpm install:cron <automation-name>
pnpm list:cron
pnpm uninstall:cron <automation-name>

Cron logs go to workspace/data/cron-<automation>.log or $JUMPYGOATHQ_HOME/data/cron-<automation>.log.

Assigned task dispatch

Create boards/tasks in the web UI or under workspace/boards/. Tasks with status: ready and a valid assignee can be claimed by the dispatcher:

pnpm dispatch:tasks           # claims one ready task
pnpm dispatch:tasks --limit=3 # optional small local batch

Install an explicit task heartbeat cron when you want periodic dispatch:

pnpm install:task-cron
pnpm install:task-cron -- --schedule="*/30 * * * *" --limit=2
pnpm list:task-cron
pnpm uninstall:task-cron

Model profiles

Agents and automations can use either direct Pi model selectors or local semantic profile keys such as fast or super-smart. Configure profiles in workspace/settings.yml or through /settings:

defaultModelProfile: fast
modelProfiles:
  fast: "provider:fast-model"
  super-smart:
    selector: "provider:smart-model"
    label: "Super smart"

jumpyGoatHq resolves profile keys before invoking Pi and stores requested/resolved model metadata on each run. Pi still owns provider auth, API keys, custom providers, and concrete model availability. Do not put secrets in settings.yml.

More docs

• Product north star: docs/vision/strategy/agent.md
• Target spec: tasks/spec.md
• Architecture: docs/ARCHITECTURE.md
• Connector setup: docs/connectors/README.md
• Private workspace Git versioning: docs/WORKSPACE_GIT.md
• Web UI package notes: packages/web/DOCS.md
• End-to-end agent testing: docs/testing/end-to-end-agent.md

What this is not

• Not a workflow builder or DAG editor.
• Not a broad personal-assistant clone.
• Not a hosted SaaS or multi-user RBAC product.
• Not a custom LLM/tool loop.
• Not a generic repo-wide chat surface.

The goal is the smallest useful open-source agent operations layer: strong primitives, limited features, and clear extension seams.