One record, many views
Every view, report and Agent operation reads the same canonical records. If two screens could disagree about a status or a count, the design is wrong.
About
We build the delivery control plane for teams, clients and AI Agents
ProjectHubs is for software teams that deliver projects for clients, keep their code in Git, and are starting to use AI Agents in delivery. It is pre-release, and the release order below carries no dates.
Why it exists
Project truth fragments across tickets, task lists, chat, schedules, files and code hosts. Priorities stop matching between views, project health becomes easy to overstate, and client decisions lose their trail.
AI Agents make this sharper. Local agent files and conversation memory are not a record anyone can audit, and a passing test report is not an approval. ProjectHubs keeps one canonical record of delivery and lets Agents do real work without taking on the human authority to approve, merge, deploy or close.
-
R0–R2
Human and client core: projects, the canonical Work Item, client governance and reports. Partly available, partly in development.
-
R3–R4
Governed Agent planning and bounded Agent execution, with human approval gates and evidence review. In development.
-
R5–R6
Schedule and workload, then integrations and decision support led by recorded demand. In development.
How we build
Humans decide, at named points
Approving a specification or plan, accepting an implementation and closing a ticket belong to authenticated people with the right project role. Agents propose; they never approve.
A control plane, not a runtime
Agents run in your own agent client. We hold no model credential, supply no compute and run no model, and Git stays the source of truth for your code.
Status from evidence
We describe a capability as available only when our delivery records show it exists. Everything else is called in development or planned, and we give no dates.
Want to follow the build?
Talk to us about your delivery, or read how the Agent protocol is designed.