Get started
Explore Crewship
GitHubGet started

From the founder

Your work should stay yours.

Your history, your agents’ memory and the AI they run on. I am building Crewship so all of it stays with the person who runs it.

Pavel SrbaFounder of Crewship
Read the story Star on GitHub

The story in seven chapters

01 / 07

The sea

Almost every app I work with keeps my history somewhere else: the chats, the decisions, and now the memory my AI agents build. If I leave, it is gone. That is the sea we all swim in.

  • A sea of appsEach one great at one job, none of them yours.
  • In someone else’s cloudYour history sits on servers you do not run.
02 / 07

Keep it yours

I think what an agent learns should belong to the person who runs it. So Crewship keeps the work, the history and the memory on your own hardware, and you can take it with you. Cloud models still receive the context you send.

  • On your hardwareConversations, memory, journal, keys and backups.
  • Take it with youAgent memory exports as plain markdown files (beta).
On your hardware
  • Conversationskept
  • Agent memorykept
  • Journalkept
  • Credentialskept
  • Encrypted backupskept
memory/ AGENT.md · CREW.md · topics/ plain markdown files you can take anywhere
03 / 07

A workshop for every crew

Every crew gets a workshop of its own: a small Linux computer with the tools and logins the job needs. The agents in the crew share it.

  • Its own workshopExactly the tools the job needs, and nothing more.
  • Its own loginsKeys assigned to the crew or to one agent, sealed in the vault.
crew · Releasedevcontainers/base:bookworm
RileyClaude Code · lead
SamCodex CLI
ContainsgitghNode 22Python 3.12Terraform 1.9bash
sidecar 127.0.0.1:9119egress allowlistread-only root
04 / 07

Work that runs itself

The work that repeats should not need you. Routines run it on a schedule, from a webhook or with one click. A run waits for a person where it matters and survives a restart.

  • RoutinesSteps, agents and people in one flow.
  • You keep the final sayImportant runs wait in your inbox.
05 / 07

Made together

Next come artifacts: documents, pages and slides that people and agents edit together, with every version kept. Documents are in nightly builds behind a switch that is off by default. Pages are in beta; slides and editing side by side are in development. I think this is the future of how we work.

  • ArtifactsDocs, pages and slides, edited together.
  • Agents in the same fileAn agent writes into the document while you watch.
06 / 07

Local AI

I believe in local AI. Bring a supported provider, or run a local model through OpenCode and Ollama. Over time, more of the work will run fully local.

  • Your runtimeClaude Code, Codex, Gemini, OpenCode or Cursor.
  • Local inferenceOpenCode with Ollama, no API key. Connected tools may still send data.
  • Claude CodeProduction-tested
  • Codex CLIWired end to end
  • Gemini CLIWired end to end
  • OpenCodeWired · runs Ollama locally
  • Cursor CLIWired end to end
Agent · Avery
RuntimeOpenCode
ModelOllama · qwen3-coder
API keyNone needed
InferenceStays on this machine
07 / 07

Built in the open

The ambition is big: one platform that can replace the rigid systems we bolt together today. For now I build it alone, and it is open source. I welcome collaborators.

  • Open sourceApache-2.0, in open beta, one release at a time.
  • A note from meThank you for reading this far.
crewship-ai/crewship

Self-hosted platform for people and AI agents.

Apache-2.0Open betaGo

“Thank you for reading this far. Crewship is the thing I wish I had, and I am building it in the open.”

Pavel SrbaFounder of Crewship

That is the story.
If it makes sense to you, support it on GitHub.

A star helps a solo, open source project more than you would think. An issue, an idea or a feature helps even more.

Star Crewship on GitHub See what it already does