Projects / 02
HiCaptain
A team workspace with human-reviewed AI assistance.
- My role
- V1 backend · V2 sole developer
- Scope
- Projects · Integrations · AI assistance
- Built with
- Next.js · Spring Boot · MongoDB · LangGraph · Docker
- Platform
- Self-hostable web app
- Project site
- hicaptain.dev · Launching soon
- Repository
- Public · MIT
Built the original team version’s backend. Developed version 2 independently, using AI-assisted implementation and review.
A note on the work,
and the choices behind it.
The problem
Project work is spread across task boards, calendars, file storage, and repositories. Adding AI creates another question: how should its context and permissions follow the people responsible for the project?
HiCaptain brings those workflows into a project-centered workspace, with organization memberships, explicit integrations, and AI-proposed changes that people can review.
My contribution
I built the backend for HiCaptain’s original team version. For version 2, I worked as the sole developer, redesigning the interface, correcting existing behavior, and extending permissions, project workflows, integrations, and AI assistance.
I used AI-assisted development to implement and review the changes, including project-scoped context, persistent memory, human approval, usage controls, and security improvements.
Permissions that follow the project
Account identity is separate from organization membership. One person can belong to several organizations with different roles, while project access depends on the active organization and the person’s assignment.
The Spring Boot API checks current membership and session versions on authenticated requests. The Next.js interface brings the project timeline, team, Kanban board, AI teammates, and GitHub activity into one view.
Connecting the tools people use
Identity sign-in is separate from explicitly connecting Google Workspace or GitHub permissions. The integrations are implemented for installations configured with their own provider credentials.
- Google Calendar and Meet for calendar reads and application-created meetings.
- Personal Google Drive folders for project files and participant sharing.
- A one-time copy of project tasks into Google Tasks.
- GitHub repository activity through the connecting user’s account, plus signature-checked webhooks.
AI work that stays reviewable
An optional Python and LangGraph worker retrieves bounded project context and produces structured planning or research proposals using OpenAI or Ollama. Authorized teammates approve the proposed tasks and decisions before they enter the project and its shared memory.
MongoDB-backed workflow checkpoints preserve saved output, while approval uses deterministic task and memory IDs to avoid creating the same records twice. Token allowances and current permission checks keep execution tied to the project’s configuration.
Self-hosting and the next release
The backend has a Java 17 Docker build and a Compose setup for MongoDB and the optional worker. GitHub Actions runs frontend, backend, and worker checks.
Version 2 is being prepared for hicaptain.dev. The source linked here is the revamped branch; the project site is launching soon.