An operating system for people who build software.
A dev-first SaaS platform where freelancers, consultants and small software houses keep clients, projects, requirements, planning and the client portal in one place — and, in the same move, serve that context to AI agents. It's where my personal engineering work is focused right now.
Independent software delivery means the project lives scattered across WhatsApp, Trello, GitHub, Notion, Jira and spreadsheets. Requirements lose their owner, scope creeps unrecorded, the client sees no progress at all — and the AI now writing much of the code arrives with no sense of what the project is. BuildOS attacks that fragmentation on the premise that structured project history is worth as much to the agent as it is to the person.
An MCP server inside the monolith exposes project context and a curated set of write operations to agents like Claude, Claude Code and Cursor — reusing the Actions and Policies the app already relies on, so an agent obeys exactly the same authorization rules a person does. It sat far out on the roadmap and was pulled forward, because it's what separates BuildOS from yet another project tracker.
Captured from the running product.
One workspace per account, clients isolated from one another, and portal access by invitation.
Keeps separate engagements from bleeding into each other — each client sees only their own work, and the invitation is what opens that door.
Status dashboard, with methodology chosen per project: sprint, kanban or simple.
Lets each project run at the cadence it actually needs — sprint when the commitment is fixed, kanban when work is pulled, simple when it's just a list with a deadline.
A flat, typed list — functional, business rule, non-functional — with estimates and scope freezing.
Records what was agreed and holds up the later conversation: once scope is frozen, every new request lands flagged OUT OF SCOPE, with a date.
Sprints, backlog, kanban board, subtasks and estimated-versus-actual time tracking.
Turns requirements into executable work and shows where estimates are slipping, task by task — the input you're missing when pricing the next project.
An update timeline with comments, progress and deadline for each project.
Lets the client follow along without chasing you for status — and keeps the conversation recorded next to the project instead of lost in a chat thread.
Its own entity, with a separate board, triage (confirm, dismiss with a reason, duplicate), severity and a post-mortem on close.
Keeps a defect from running over the sprint commitment: it lands on a separate board, gets triaged before becoming work, and the client can report it straight from the portal.
An MCP server inside the monolith, connected through the browser (OAuth) or with a token.
Lets an AI agent work knowing what the project actually is — it reads requirements, tasks and roadmap, and writes back under the same Policies a human user obeys.