Build Your Team...
without hiring one.
With Launchpad Studio, thrive as a team of one. Create mini-experts with their own memory and skills — delegating to each other, working while you don't, sharpening with every use. Open source & runs on your machine.
with Agent Profiles — Give agents their own skills, workflows, and memory — reliable experts you can call on by name.
Free to run. On your own machine
Connect any model provider with your own API key or CLI plan.
Follow along on GitHubGets better with use
Threads are constantly distilled to generate candidates for memories and skills.
- Thread memories: Focused memory for the task/topic - thread at hand. These become candidates for Agent and Project Memories.
- Project memories: Scoped to the working directory or git repo. Different Agents contribute to and consume them across threads and over time.
- Agent memories: Memories deemed to be relevant to the agent across threads and projects.
- Global memories: Shared across every agent and every project.
Nothing lands silently
A reflection pass proposes candidate memories and skills. You accept, edit, or reject each one — and can undo a write later. Always a human in the loop.
Skills distilled from repetition
When an agent repeats a procedure often enough, it's generalized into a reusable Skill — shipped parked and disabled until you approve it.
Climb the autonomy ladder
Hand over as little or as much as you trust an agent with. Each rung does more on its own than the last.
Agents & profiles
A specialist you shape once and reuse — its own persona, curated skills and tools, memory, and model/provider — instead of one generalist you re-brief every time.
Address books & delegation
Agents hand work to each other through a curated address book. Mix and match model sizes and providers to extract the most out of your agents. Larger models can be used for the heavy thinking, and hand off grunt work to smaller, faster, and cheaper models.
Tasklists
Multi-step work gets broken into steps, each run by a fresh, focused agent instance — the structured replacement for a hand-rolled "keep looping until done" prompt - replaced the Ralph loop for us
Projects
Hand over a whole goal. An agent interviews you to capture intent, drives itself through tasklists, and can only call it done once an independent verifier agrees.
Assignments
The top rung — agents that start themselves, on a cron schedule, a webhook, or when a connected data source changes. No one has to ask. Hand off work to your daily drivers so that your changing preferences are incorporated immediately without requiring you to update this elsewhere