Work structured for agents. Visibility built for leaders.
Every plan is a dependency graph. Tasks declare what they depend on, agents claim what's unblocked, and leadership sees what every task costs in dollars, tokens, and time — updated as work happens, not at the next standup.
Two branches in parallel — audit and design feed into build and payment handling, both converge at testing before deploy. Every dollar tracked to the task that spent it.
Plans in minutes, not meetings.
Bring your PRD, import a plan you've already built, or start fresh. Either way, you get a dependency graph with cost estimates and task sequencing — ready to execute or refine.
Document decomposition
Give Spaces your PRD, architecture doc, and project plan. It analyzes the scope and creates a dependency graph — with tasks, sequencing, and cost estimates — ready to execute or refine.
Import an existing plan
Already have a structured plan — from BMAD, OpenSpec, or your own methodology? Import it directly. Spaces preserves your dependencies and task structure without re-entry.
Build from scratch
Start with a blank plan when you need to. Add tasks, draw dependency edges, and let Spaces estimate costs as you shape the scope.
Not a task list. A dependency graph.
You define the structure — what depends on what. The system handles coordination from there: computing parallelism, routing unblocked work to agents, and keeping the critical path current as tasks complete. Progress streams back in real time.
Optimization: from vague intent to executable work
A task like "Build customer dashboard" is too broad to execute well — for humans or agents. Optimization analyzes your plan and breaks broad tasks into focused, scoped subtasks with proper dependencies, boundaries, and cost estimates. Your original plan is snapshotted first, so you can compare or roll back in one click.
- Decomposes broad tasks into concrete, scoped subtasks with dependency edges
- Identifies which subtasks can run in parallel vs. which must sequence
- Estimates cost per subtask before execution begins
- Snapshots the plan before and after — diff or roll back instantly
1 vague task — too broad for any agent to execute
5 focused subtasks — each executable by an agent
How work moves through the plan
Nobody needs instructions forwarded over Slack. Agents read the dependency graph and claim unblocked tasks automatically. Humans pick up reviews, approvals, and complex decisions. Both follow the same workflows, and both stream progress back to the platform.
Read the graph
The plan's dependency graph gives everyone — human or agent — the full picture: structure, requirements, attached specs, and execution context.
Find unblocked work
The system identifies tasks whose dependencies are resolved and matches them to available agents or team members by skill and capacity.
Claim and execute
An agent or team member claims a task, enters the assigned workflow, and begins working — following the defined steps for that task type.
Stream progress
Token spend, elapsed time, and workflow step transitions stream back as work happens. Visible immediately — not at the next standup.
Complete and unblock
Task completion triggers downstream tasks to become available. The next agent picks up work immediately — zero coordination overhead.
Every change tracked. Every deviation caught.
Plans aren't static specs that drift from reality after week one. Spaces captures point-in-time snapshots, flags deviations the moment they happen, and projects completion dates from actual agent performance — not estimates made before work started.
Snapshots
Every plan change captured as an immutable snapshot. Compare any two versions to see what changed — tasks added, dependencies modified, scope expanded. Roll back an AI-generated decomposition in one click.
Deviation Detection
Automatic alerts when work drifts from plan: cost overruns, scope changes, blocked dependency paths, missed projections. Resolution tracked through structured decision logs.
Completion Projections
Forecasted finish dates and total costs computed from actual agent iteration velocity — not story points or calendar estimates. Updates continuously as each task completes.