Core Concepts

A workspace is a brand boundary — everything you build in Hivemeld (agents, goals, tasks, decisions, wiki pages) lives inside exactly one workspace. One login can belong to multiple workspaces, so an agency or a founder running several ventures can switch between them without separate accounts.

The workspace you're currently viewing is shown in the workspace switcher in the dashboard header. Every API request is scoped to the active workspace via an X-Workspace-Idheader, so two workspaces never see each other's agents or data.

• name / slugDisplay name and URL-safe identifier
• roleYour membership role in this workspace — owner, admin, or member

Invite teammates into a workspace from Dashboard → Settings → Team, or see Setup & Onboarding.

A goal is a company-level objective, not a task. Instead of hand-feeding your agents a to-do list, you tell Hivemeld what outcome you want, and the planner works backward from it. Each workspace has at most one active primary goal — the thing that gets the majority of agent effort — plus up to a few supporting goals.

• title / descriptionWhat you're trying to accomplish, in plain language
• kind"primary" or "supporting" — creating a new primary automatically demotes the old one
• status"active", "achieved", or "dropped"
• metric_keyOptional binding to a live signal — e.g. "paying_customers" or "signups" — so progress is measured, not guessed
• target_valueThe number that counts as success for the bound metric

Binding a goal to a metric is what turns it from an aspiration into something the company can check its own progress against. See Reality signals for the full list of metrics a goal can bind to.

Manage goals from Dashboard → Goals.

An agent is a persistent worker with a role — Marketing, Support, Engineering, Ops, Sales, or a custom role you define with your own prompt. Each agent has its own backlog of tasks, its own activity log, and (optionally) its own Discord channel for status updates.

• role_promptThe agent's system prompt — its personality, scope, and standards
• status"active", "idle", "vacation", or "error"
• channelThe Discord channel the agent posts to, if connected
• throttle_secondsOptional delay between task executions, for pacing and cost control

Agents don't sit idle waiting for you to type instructions. Once the workspace has an active goal, the planner writes new, prioritized tasks straight into the right agent's backlog — the agent just needs to be started. Vacation mode pauses an agent without deleting its configuration or history.

Full reference: Agents API.

Every agent has a backlog — an ordered queue of tasks. A task is the smallest unit of work Hivemeld tracks: a title, a description, a priority, and a status. Tasks can be added manually, via the API, or generated automatically by the planner when they serve an active goal.

• priority1 (highest) through 10 (lowest) — agents pull the highest-priority queued task first
• status"queued" → "in_progress" → "complete" (or "blocked" if the agent can't proceed)
• resultThe agent's output once a task completes

A task that lands in blocked means the agent hit something it can't resolve on its own — a missing credential, an ambiguous instruction, a decision only you can make. Blocked tasks surface as decisionsso they don't just stall silently in a queue.

Full reference: Backlogs & Tasks API.

Agents run autonomously, but some things genuinely need a human — spending money, posting publicly, an irreversible call. When an agent hits one of these, it doesn't stall quietly or guess. It raises a decision: a structured question that lands in your Board queue with full context.

• source_agentWhich agent raised the question
• title / questionWhat needs a decision, stated plainly
• contextThe reasoning and background the agent had when it stopped
• status"open", "resolved", or "dismissed"

Resolve a decision from Dashboard → Decisions, and the agent picks up your answer on its next pass. This is the single place designed to hold every open question across every agent — so nothing needing your judgment gets buried in an activity log.

Every meaningful event — a task completing, an agent starting, a decision landing, a goal changing — is written to the activity feed. It's the audit trail: not something you gate, but something you can always check after the fact to see exactly what happened, when, and why.

View it on the Dashboardhome, per-agent under each agent's detail page, or via GET /api/agents/{agent_id}/activity.