No Workspace Selected
Create or select a workspace to manage access and runtime credentials.
Goals
The objectives the workforce pursues. Org goals are the team mission (every member inherits them); member goals are assigned to one worker. A goal can auto-close when its measurement target is met, or be closed by hand.
Workforce
Your AI team. Each member holds a role, pursues a mandate (its goals), and can be made proactive with triggers that spawn tasks on a schedule. Select a member to see its mandate and triggers, or hire a new one.
Track record — cost & outcomes per role
What each role has cost and how it's performed across recent runs — the prior a cost-informed hire reads.
Team templates — start from a curated team
A template seeds specialist roles (activated, ready to hire against) plus the starter org goals they exist to pursue — the one-click cold start. Re-applying is safe: goals already present are skipped.
Activity
Recent runs in this workspace — what the team (and human tasks) are doing, who is acting, toward which goal, and at what token cost. Select a run for its detail. The Conversations view is the register of goal-planning chat sessions.
Inbox
Everything waiting on you. Answer an agent's clarification to resume its run; runs that ran out of budget are listed so nothing stalls silently; and the learning loop's promotion proposals queue here for your approval before any config goes live. This is how you stay in the loop without babysitting the team.
Resources
Where the tokens go, against what was allocated — burn by member, goal, and role, with the per-member and per-goal budgets the engine enforces at every spawn. Set or edit an allocation inline; remaining headroom drives the near-cap and exhausted alerts.
Daily burn
Allocations — most pressured first
Members — burn vs. allocation
Goals — burn vs. allocation
Roles — burn
Learning
The workforce improving itself, under your control. Underperforming roles and workflows are flagged from real outcomes; an LLM proposes a revised candidate as a DRAFT; the scoring harness replays history against the incumbent and gates promotion. When the autonomous steward wins the gate, the promotion waits here (and in the Inbox) for your approval — nothing goes live without you. Every run and every approval is recorded.
New API key secret
Ontology Profiles
Extracted Memory
The entities, relations, and facts the memory pipeline has extracted for this workspace's shared (teamspace) memory. Use this to confirm the assistant's memory is building.
Entities
Relations
Facts
Advanced: Debug Views & Projections
Debug Views
No view selected
Use this tab to inspect and debug persisted memory projections and query behavior.
Parameters
Consumers
Advanced Governance
Governance Audit Rollups
View Query Runner
Saved Query Presets
End-to-End Workflow Runner
Agent Refinement Diagnostics
Run Diagnostics
Rollups
Agent Planner Diagnostics
Run Diagnostics
Rollups
Agent Critique Diagnostics
Run Diagnostics
Rollups
Embedding Diagnostics
Rollups
Cross-Diagnostics Dashboard
Cross Rollups
Dashboard Presets
Roles
The positions members hold: an instruction overlay plus advisory priors (preferred agents as skills, preferred workflows) and memory-access bounds. Save stores a DRAFT; Activate makes the version live (one ACTIVE per role id). Skills and workflows bias the orchestrator — they are not a whitelist.
Agents
No agent selected
Capabilities
This catalog lists the agents currently resolvable for runs (built-ins plus your active workspace agents). Author and version workspace agents below.
Workspace agent definitions
Author manifest-only agents. Save stores a DRAFT; Activate makes the version live (one ACTIVE per agent name); Retire removes it.
Workflow definitions
A workflow declares an orchestrator agent, a dispatchable-agent whitelist, and hard constraints — never a fixed sequence. Save stores a DRAFT; Activate makes the version live (one ACTIVE per workflow id). Activation requires every referenced agent and the prompt template to resolve.
Memory pipeline definitions
A memory pipeline declares a memory-orchestrator agent, a dispatchable-stage whitelist, and hard constraints — never a fixed sequence. Save stores a DRAFT; Activate makes the version live (one ACTIVE per pipeline id). Activation requires every referenced stage agent and the prompt template to resolve.