Every project in HEXi has its own dedicated conversational agent. Where the Orchestrator operates at guild level, the Project Agent lives inside a single project — deeply familiar with its milestones, sprints, team, and history. Ask it anything about the project and it will answer from that context, not from a generic model.
The Project Agent is pre-loaded with a full snapshot of its project on every conversation — milestones, sprints, sprint items, team members, and recent decisions. You never have to re-orient it. You just ask.
Conversational project intelligence
Ask in plain language: "What's blocked?", "Who owns the auth milestone?", "Which sprint items are overdue?", "Summarise last sprint's decisions." The agent answers from live project data, not a cached summary.
Sprint and item management
The agent can create and update sprints, sprint items, and milestones — and reassign items to team members — all through conversation. It understands the project hierarchy (Project → Milestone → Sprint → Sprint Item) and keeps everything consistent.
Work log management
Log, query, and update work sessions against any sprint item. Each work log captures duration in minutes, notes, date and time, and a status — Started, Complete, Billed, or Cancelled. The agent can produce a time summary for any item, sprint, or actor on demand.
Sprint item triggers
Create automation rules that chain sprint items together. When item A reaches a given status, trigger an action on item B — update its status, assign it, or fire a custom payload. The agent can list, create, and delete triggers through conversation.
Project decisions
Raise a governance decision scoped to the project directly from the project chat. The agent initiates the decision, routes it through the relevant council agents, and records the outcome against the project's decision history.
Knowledge graph integration
The agent can snapshot the entire project hierarchy into the knowledge graph with a single command — writing deduplicated nodes and relationships for every milestone, sprint, item, and team member. It can then search the graph to answer questions that span projects, actors, and decisions.
When you open a project chat, the server fetches the full project snapshot — milestones, sprints, active sprint items, team members, and recent decisions — and injects it into the agent's context before your first message is sent. The agent knows where the project stands before you say a word.
Thread history is persisted in PostgreSQL with vector recall, so returning to an existing conversation picks up exactly where you left off. Quick-action chips on the chat surface surface the most common project questions with a single click.
Scoped to one project. Each Project Agent is instantiated with a specific projectId. It cannot read or modify other projects. Guild-level operations — cross-project decisions, agent reputation, global knowledge — remain with the Orchestrator.
Access-controlled. All project API routes enforce project membership. Only team members with access to a project can open a project chat or interact with its agent.
Knowledge proposals, not direct writes. When the agent identifies knowledge worth capturing, it queues an episode for human review rather than writing directly to the graph. The Librarian reviews and approves it — keeping the knowledge graph governed, not agent-written.
Project reading
Get project, list milestones, sprints, items, decisions, files
Project writing
Create/update sprints, items, milestones; reassign items; update status
Work logs
List, add, and update work log sessions with duration, notes, and status
Triggers
List, create, and delete sprint item automation triggers
Governance
Initiate project-scoped decisions routed through the council
Knowledge graph
Snapshot project, search facts and nodes, queue episodes, add triplets
Private notes
note-to-self, search-notes, publish-note
Workspace
Read and write files scoped to the project workspace directory
Start managing your project with an agent
Every project created in HEXi gets its own agent. Open a project, start a conversation, and your agent is ready — oriented, informed, and able to act.