MANO 1.1 AGI roadmap

Choosing and running the right tool for a step instead of answering from memory alone.

Where MANO is

MANO picks its own tool each step from a whitelist: knowledge search, deep sub-reasoning, note-taking, finish.

Next step

Add code execution and web retrieval as first-class mission tools.

Breaking a goal into ordered steps and adapting the plan as results come in.

Where MANO is

Missions run a bounded plan-act-observe loop (up to 8 steps) with a deep plan stage before drafting.

Next step

Multi-session plans that survive across days and resume where they stopped.

Attacking its own answer before delivering it, and scoring the result against the goal.

Where MANO is

Every deep run has an adversary pass plus synthesis; missions self-score 0-100 against the goal.

Next step

Executable checks (tests, builds) as the scoring signal instead of self-judgement.

Recalling relevant facts and past work rather than starting cold each time.

Where MANO is

Vector knowledge memory feeds chat and missions; missions store durable lessons per user.

Next step

Automatic consolidation and forgetting so memory stays sharp as it grows.

Turning its own past work into rules that change future behaviour.

Where MANO is

Missions and lessons compile into an operating doctrine that loads back into MANO's context.

Next step

Score-weighted training and fine-tuned weights served behind the same mano-1.1 ID.

Deciding what to work on first when several goals compete.

Where MANO is

Scheduled agents run goals on a timetable; ordering within a mission is model-chosen.

Next step

A standing goal queue with value, urgency, and cost ranking across missions.

Knowing its own limits, cost, and confidence — not consciousness.

Where MANO is

MANO reports depth, domain, substrate, stage timings, and refuses out-of-scope requests.

Next step

Calibrated confidence and explicit 'I cannot verify this' signals on every claim.

Starting useful work without being asked, safely and within budget.

Where MANO is

Nothing runs unprompted except schedules the user creates.

Next step

Opt-in background missions with hard spend caps and an approval log.