Capabilities
Everything an AI workforce needs to run on its own.
Five capability areas, one connected system. Each feature makes the others more useful: dispatch feeds memory, memory improves plans, and plans direct the fleet.
Execution
Tasks leave the planner and land on real machines, with the right model, in isolation, with a way back to you.
Cross-machine dispatch
Send a task to a named machine or let the coordinator place it. Every worker starts in its own git worktree, so parallel agents never collide.
Difficulty-aware model routing
Hard work and review go to the strongest model. Mechanical changes go to faster ones, and bulk goes to Codex on a separate pool.
Verification gate
Definition-of-done audits and independent verify passes run before anything is accepted or merged.
Interventions
Agents ask instead of guessing. The question reaches the phone with options, and the answer resumes the worker.
Session recycling
Long sessions compact, hand off and resume cleanly instead of degrading silently as context fills up.
Planning
Projects, sprints and briefs give the fleet something durable to execute against, and give you the day at a glance.
Projects & sprints
Goals, tasks, dependencies and acceptance criteria that persist across sessions, machines and models.
Morning & evening briefs
A CEO-style brief that combines priorities, calendar, email signals and overnight fleet output, and adapts to feedback.
Planner
Today’s plan built from priorities, momentum and context, with explained ordering.
Scheduled project reviews
Stakeholder-ready reports generated on a schedule from what actually shipped.
Reminders
Time-based nudges delivered to the channel you’re already in.
Intelligence
Everything the system learns is compiled, cited and reused, so completed work makes future work cheaper.
The Brain
A compiled wiki of projects, decisions, concepts and people with provenance, contradiction flags and a review queue.
Project memory
Search per project or across every project for the fix, decision or pattern that already exists.
Lessons
Captured at handoff, reviewed, graduated and attached to future dispatches automatically.
Absorb
YouTube talks, articles and X posts become sources the Brain can cite in answers and plans.
Grounded answers
Chat, Telegram and voice answers come from compiled knowledge and cite the pages they used.
Reach
The same record of progress that runs the fleet feeds content and relationships.
Content engine
Topic discovery and grounded drafts built from real project progress, decisions and lessons.
X distribution
Timeline sync and post performance feed back into what to write next.
Network radar
A relationship engine that ranks who to reply to, follow up with and reconnect with.
Operations
The system watches its own capacity, health and efficiency, and uses that to make better routing decisions.
Usage observatory
Claude and Codex subscription usage is probed continuously, paced against weekly limits, and shown per account.
Metrics
Sessions, dispatches, tokens per model, cache hits, merged PRs and interventions, rolled up hourly.
Health & activity
A cross-project timeline and health view for every service and machine.
Device plane
Agents on any machine can drive attached phones and desktops through MCP tools.
Per-project accounts
Client projects can run on their own model subscription, isolated from everything else.