OpenCrane is building a company workspace where every employee has a private AI assistant and teams can delegate repeatable work to shared agents. Assistants should understand how people work, use the right company tools and knowledge, and carry useful context between conversations.
The organisation chooses where OpenCrane runs, which models it uses, what each assistant may access, and which actions need a person's approval. Conversations, files and company configuration stay under its control. Data sent to an external model provider or integration follows the organisation's chosen configuration.
OpenCrane brings five things together:
- People and teams: company membership and permissions determine who may use or manage each resource.
- Assistants: a personal assistant works privately for one employee; a shared agent is intended to carry out a defined company task with its own permissions.
- Conversations: a place to ask for work, follow progress and return to its history. An assistant and a conversation are different things: the assistant helps; the conversation records the work.
- Tools, knowledge and files: tools act in other systems, knowledge supplies context, and files hold inputs and results.
- Company controls: administrators manage access, model providers, spending limits and activity.
The intended experience is straightforward: ask for help preparing a customer meeting, let the assistant use the company information it may access, review any proposed action, and return later to the same work. That complete journey is the product goal; its individual capabilities have different implementation states.
OpenCrane is pre-MVP. People can complete setup, get personal-assistant answers, talk in a group, and ask a company assistant for help in a linked chat. They can review and edit its answer before returning it to the group. These text-conversation journeys have passed live testing in the 0.11 review baseline, including saved answers after a browser reload.
The wider vision still needs work: assistants do not yet perform governed tool actions, remember information across conversations, run shared schedules or delegate autonomously to other assistants. Administration and recovery have their own remaining product and qualification work.
See what is built and what remains for the current distinction, CHANGELOG.md for capability changes, and plan.md for active work.
The web workspace talks to an OpenCrane server that checks permissions and saves the work. When an assistant needs to run, it uses an isolated conversation computer. The computer can stop and be replaced while the conversation and saved workspace remain recoverable. Models, integrations, memory and files have their own shared services.
The architecture overview maps these responsibilities to the current applications and stores.
If your organisation already runs OpenCrane, sign in at its address and follow the personal-assistant setup guide.
To install a development instance, start with the installation guide. It links the supported Kubernetes setup, identity configuration and required deployment inputs. The 0.11 release installation uses a fresh database baseline.
For local development:
npm ci
npm exec nx show projectsUse focused Nx build and test tasks for the part you are changing. Contributor instructions start in AGENTS.md; the contributor guide explains the development workflow.
| Path | Purpose |
|---|---|
| apps/ | Deployable applications and installation tooling |
| libs/ | Reusable product capabilities and shared infrastructure |
| libs/contracts/ | Shared API and runtime contracts |
| website/ | User, operator and integrator documentation |
| docs/adr/ | Architecture decisions |
OpenCrane is available under the GNU Affero General Public License v3.0 or later.