AI agents
A specialised AI workforce that proposes — and never silently decides
Agents in Project Planner have roles, tools, skills, MCP connections, memory scope and permissions. They produce artifacts and findings for you to review. Destructive actions always require confirmation, and a run is never shown as model-backed unless a provider is actually configured.
Agent groups
Five groups, nineteen roles
Software
Architecture, frontend, backend, database and API specialists that turn requirements into components, contracts and schemas.
- Software Architect
Owns system boundaries, component contracts and technology decisions.
- Frontend Engineer
Plans screens, state and offline behaviour for the clients.
- Backend Engineer
Designs domain services, workers and event handling.
- Database Engineer
Designs schema, indexes, migrations and sync log.
- API Engineer
Specifies endpoints, pagination, errors and versioning.
Design
UI and UX designers and an accessibility reviewer that produce specifications and check flows against WCAG.
- UI Designer
Produces UI specifications and component inventories.
- UX Designer
Maps user flows and validates task completion paths.
- Accessibility Reviewer
Checks contrast, keyboard paths and semantics.
Planning
Product planner, requirements analyst and technical planner that scope, trace and sequence the work.
- Product Planner
Turns the idea into scoped phases and milestones.
- Requirements Analyst
Extracts, de-duplicates and traces requirements.
- Technical Planner
Estimates tasks and sequences dependencies.
Quality
Error, testing, security, compliance and performance agents that audit with evidence and derive test plans.
- Error Analyst
Classifies failures and proposes remediations.
- Testing Agent
Derives test plans from acceptance criteria.
- Security Agent
Reviews data flows, secrets and permissions.
- Compliance Agent
Checks privacy, consent and accessibility obligations.
- Performance Agent
Flags hot paths and sets budgets.
Documentation
Documentation, blueprint and change-log agents that keep the written record current and generate the final blueprint.
- Documentation Agent
Keeps architecture and setup docs current.
- Blueprint Agent
Assembles the end-to-end project blueprint.
- Change Log Agent
Summarises what changed and why.
Agents
Preview — demo project contentSoftware
- Software ArchitectIdle
- Frontend EngineerRunning
- Backend EngineerWaiting
- Database EngineerIdle
- API EngineerQueued
Design
- UI DesignerIdle
- UX DesignerIdle
- Accessibility ReviewerRunning
Planning
- Product PlannerIdle
- Requirements AnalystIdle
- Technical PlannerIdle
Quality
- Error AnalystIdle
- Testing AgentIdle
- Security AgentRunning
- Compliance AgentIdle
- Performance AgentIdle
Documentation
- Documentation AgentIdle
- Blueprint AgentIdle
- Change Log AgentIdle
Agent activity
A live HUD for every run
Agent Activity
Preview — demo project contentRunning
Waiting
Queued
Completed
| Agent | Task | Status | Current step | Progress | Elapsed | Tools |
|---|---|---|---|---|---|---|
| Frontend Engineer | TSK-004 Offline work-order store | Running | Drafting sync-state machine for pending changes | 62% | 4m 12s | 7 |
| Security Agent | Data-flow review (REQ-004) | Running | Checking classification of edges leaving the tenant boundary | 35% | 2m 03s | 3 |
| Accessibility Reviewer | TSK-006 Mobile navigation review | Running | Evaluating focus order on the work-order list | 81% | 9m 48s | 12 |
| Backend Engineer | TSK-007 Conflict-resolution strategy | Waiting | Question: prefer last-writer-wins for signatures, or block and ask the technician? | 40% | — | 5 |
| API Engineer | TSK-008 Sync API delta endpoints | Queued | Waiting for TSK-007 decision | 0% | — | 0 |
| Database Engineer | TSK-002 Schema design | Completed | Produced ART-003 Database design v2 | 100% | 11m 30s | 9 |
Runs update live from execution events. Simulated runs are labelled; a run is never shown as model-backed without a configured provider.
Execution records
Every run stores agent, task, status, start and end time, model, provider, tool calls, artifacts, warnings, tokens and estimated cost — never credentials.
Questions, not guesses
When an agent needs a decision it asks, with context, options and why it matters. Approved answers are recorded in decision memory.
Simulated runs are labelled
Without a configured provider the desktop app can demonstrate agent flow using clearly labelled simulated runs. It never pretends a model was called.
AI providers
One abstraction, any provider
| Capability | Status | Notes |
|---|---|---|
| Provider and model selection per agent | Planned | Configured in AI Providers and per-agent settings |
| Fallback providers | Planned | Primary → Fallback 1 → Fallback 2; each switch is logged |
| Token and cost tracking | Planned | Per run, per provider, per project; shown in Cost Analysis |
| Request history | Planned | Redacted request metadata, never prompt secrets |
| Provider health | Planned | Connected, disconnected, error, not configured |
MCP, skills and memory
Tools and knowledge, kept explicit
MCP servers
Connected servers, available tools, status, permissions (read, write, network, execute), configuration, errors and usage. Add, remove and configure connections; assign servers to agents.
Skills
Versioned instruction sets with a source, capabilities and required tools. Enable per project and assign to agents. Skills are reusable across projects.
Memory you can read
Project memory, decision memory, user preferences, agent memory and technical memory are separate, inspectable and editable. Each item records its source and confidence. No hidden state.
Preview content