Observe
Watches live sessions for struggle signals — project turns, recent tool errors, wake reasons, project changes. The work surfaces before it stalls out.
Operator → Morpheus
Observe in real time.
Find weak agents. Fix the system.
Morpheus is the autonomous mode behind Operator. It watches live sessions for struggle signals, reacts to completions, failures, schedules, and your messages, then creates specialists, verifies the work, and rewrites weak prompts so the system gets better over time.
Screenshot needed
Ideally a 6-8s loop showing the cycle: Observe (counters tick: turns, errors) → Find (an agent's score updates, marked 'learning') → Fix (specialist created, verification passes, prompt revised). Each phase highlights as it runs, then loops. Static fallback: three-column dashboard with the same Observe / Find / Fix panels at rest.
/shots/morpheus-loop.pngThe Loop
Three phases. One feedback cycle.
Morpheus is not a separate AI. It is what Operator does when nobody is typing — watching, evaluating, and improving the agents that run inside your Spaces.
Find
Scores dynamic specialists over time. Tells you when an agent is reliable, when it is learning, and when a better one should be created.
Fix
Auto-discovers Grasps, spawns specialists for failing tools, verifies output before reporting success, and rewrites weak prompts so the next run is sharper.
Wake events
Four ways Morpheus comes alive.
It does not need a prompt to act. It reacts to events the system already produces.
Task completion
A run finishes — Morpheus decides whether to verify, ship, or queue what comes next.
Tool failure
An action errors out — Morpheus catches it, retries with a different model or specialist, and reports back.
Schedules
Cron-style or one-shot. The Space wakes itself at a fixed time to run a brief, a sweep, or a sync.
Your messages
You write a line. Morpheus picks it up, decides if it can act, and either runs or asks a clarifying question.
Policies
Three levels of autonomy.
Pick per Space. Tighten when you don't trust it yet, loosen when it earns it.
Observe
Watch only. Useful for shadowing a workflow before granting execute rights.
Local Execute
Run actions inside the Space. No outbound writes to other systems without approval.
Execute & Commit
Full autonomy within declared boundaries — write to the Graph, ship the change, close the loop.
Guarantees
What Morpheus won't do.
Autonomy without trust is theatre. These are the floor Morpheus runs on top of.
Verification before success
Morpheus does not call a task done until a Reviewer specialist confirms the result. No silent failures dressed up as wins.
Specialists persist
Agents that work survive the session. The next run starts with what already proved itself, not from zero.
Prompt rewrites are scored
A rewritten prompt has to beat the old one on real tasks before it replaces it. No regressions, just measured improvement.
Bounded by your policies
Morpheus runs inside the autonomy ceiling you set per Space. It cannot quietly upgrade itself.