Subsystem directory
61 declared subsystems, with their actual tasks, gates and recorded evidence.
A declared task state is a planning record. Imported or historical evidence does not prove that a service is healthy today. Open a subsystem to inspect its individual gates.
| Subsystem | Declared state | Tasks |
|---|---|---|
| S01 · Brain / Core Intelligence | built, not well structured | 13 |
| S02 · LLM Integration | complete for now | 8 |
| S03 · Tool Utilization | complete for now | 13 |
| S04 · Device Control | complete for now | 9 |
| S05 · Browser Control | complete for now | 7 |
| S06 · Document Creation | half-built — reading is strong, *creation* is three ad-hoc writers | 8 |
| S07 · Session & Context Management | built, not well structured | 7 |
| S08 · Voice Enrollment | built, **measurably weak** | 6 |
| S09 · Face Enrollment | built, **never run on the current backend** | 7 |
| S10 · Voice Recognition (speaker identification) | built, not well structured | 7 |
| S11 · Face Recognition | built, not well structured | 7 |
| S12 · Multi-Device Coordination | built, not well structured | 9 |
| S13 · Notification Management | built, not well structured | 7 |
| S14 · Proactivity Engine | complete for now | 10 |
| S15 · Recommendation Engine | **removed by decision 2026-10-03** — Mushegh answers and acts; it does not suggest (D-27). Nearest neighbours stay as tools and must not grow into a recommender. (retired) | 7 |
| S16 · Task Queue & Execution | **Todoist is the only task list (2026-10-16, `736259e`)** - Notion is no longer a task store; the to-do tools, briefs, world-model goals, backlog pass and health check all read Todoist (or the local queue when no token is set). Token set and live on the VPS. | 9 |
| S17 · Morning Brief | complete for now | 7 |
| S18 · Microphone & Speaker Management | complete for now | 14 |
| S19 · Composio / MCP Integration | built, not well structured | 7 |
| S20 · External API Integrations | dark — code complete, controls nothing; blocked on credentials for the integrations that are not enabled (D-46: dark work is not progress) | 7 |
| S21 · Personal Time Management | built, not well structured | 7 |
| S22 · System Recoverability | complete for now | 8 |
| S23 · 24/7 Reachability | complete for now — **currently parked by design** | 9 |
| S24 · Knowledge & World Model | **removed 2026-10-04 (M04)** — D-26 is carried out, not just decided. `graph.py` and `tools/graphmem.py` are deleted, the `graph` lazy tool group is unregistered, the L5b recall layer is gone from `fused_recall`, the background reviewer no longer writes triples, and `what_i_dont_know` no longer speaks from a store the owner had stopped trusting. `world_model.py` survives: it is goals and events, not relations. Measured: recall p95 improved from 27.0/42.0/28.6 ms to 11.3/15.6/15.5 ms, and the advertised tool count fell to 151 (retired) | 6 |
| S25 · Personality & Interaction Style | complete for now | 7 |
| S26 · Multi-Modal Perception | the seam is built and gated, the hardware is not — mock-only evidence, no detector and no camera | 9 |
| S27 · Contextual Awareness | built, not well structured | 6 |
| S28 · Environmental / IoT Orchestration | **live 2026-10-16** - token installed, the VPS reaches Home Assistant Green over the tailnet (ACL grant tag:server -> <HA tailnet IP> tcp:80), `/api/` answers, 27 entities visible. Unproven: one real device actuated by voice and read back (28.F4), and a week of scenes (28.E1). | 11 |
| S29 · Automation & Workflow Engine | built, not well structured | 8 |
| S30 · Persistent Memory | built — **SQLite is the store by decision** (D-02); the PostgreSQL migration is cancelled and its scaffolding deleted. Vault markdown is the human source of truth (D-22). | 18 |
| S31 · Self-Monitoring, Diagnostics & Self-Healing | complete for now | 16 |
| S32 · Redundancy & Failover | **partially missing at the architecture level** — model failover is real; there is one host. Replication, promotion and split-brain prevention are deferred pending an owner decision on whether there is ever a second brain host. | 8 |
| S33 · Goal & Project Management | built, not well structured | 7 |
| S34 · Health & Wellness Monitoring | half-built — tools exist; a data source now exists (a Garmin Instinct 3 exporting raw FIT, 34.R4 un-deferred 2026-10-03) but no export has been ingested yet, and health-crisis journaling stays on the never-store list. | 8 |
| S35 · Crisis & Emergency Response | half-built — system-level crisis protocols exist, *owner*-level does not | 7 |
| S36 · Security & Access Control | complete for now | 12 |
| S37 · Privacy & Data Governance | built, unproven by use — hygiene real, policy now enforced at the write boundary | 9 |
| S38 · Communication Hub | built, not well structured | 7 |
| S39 · Multi-Agent Delegation | built, thin | 11 |
| S40 · Financial & Asset Management (delegated to RUBEN) | **delegated to RUBEN (D-97), built 2026-10-15** - `ask_ruben` briefs, runs a saved RUBEN session (follow-ups, file intake), reviews the tool log, speaks short answers whole and long ones as tone + key sentences; the full transcript goes to Telegram on request via RUBEN's own bot. Open: six-language routing fixture (R4), conversation-store half of never-store (R8), a week of real use (E2). | 15 |
| S41 · Research & Synthesis | half-built | 9 |
| S42 · Media & Entertainment Control | built, not well structured | 7 |
| S43 · Presence & Continuity | half-built | 9 |
| S44 · Ethical Reasoning, Safety & Alignment | complete for now | 9 |
| S45 · Explainability & Transparency | complete for now | 15 |
| S46 · Predictive Analytics & Forecasting | **removed by decision 2026-10-03** — no ML forecasting (D-28). A deterministic calendar look-ahead serves scheduling; that logic belongs to S21 and S33. (retired) | 7 |
| S47 · Inventory & Resource Management | **removed by decision 2026-10-03** — fully missing, and the replenishment pipeline built inside it was never used (D-42). (retired) | 7 |
| S48 · Collaborative Multi-Agent Coordination | built, thin — one-way delegation to an external router; no coordination layer. | 7 |
| S49 · Fabrication & Manufacturing Control | **dropped 2026-09-20, re-signed by the owner 2026-10-03** (D-47, D-48) — physical vanity scope belonging in a standalone slicer/CAD utility. No open tasks remain; the decision record is docs/subsystems/49-fabrication-and-manufacturing.md. (retired) | 3 |
| S50 · Legacy Continuity & Knowledge Preservation | half-built — the substrate exists, the continuity policy does not. | 7 |
| S51 · Encryption & Secret Protection | absent — no encryption at rest anywhere in the system | 4 |
| S52 · Household Chores & IoT Housekeeping | built 2026-10-16, off by default - empty-house lights-off and appliance cycles through the same gate and budget; heating setback is held for a yes | 4 |
| S53 · General Maintenance & Environment Integrity | partial 2026-10-16 - broken installs refused at deploy, drift reported daily, retention enforced per record; the real restore drill (53.R2) is open | 4 |
| S54 · Socialization & Relationship Continuity | built 2026-10-16 - greetings from real state, yesterday followed up once, small talk only on his social turns | 4 |
| S55 · Humor (integral to the brain, not a module) | built 2026-10-16 - humor in the manner note, gated, and calibrated from his reactions | 4 |
| S56 · Tactical Operations & Situational Awareness | built 2026-10-16 - `sitrep`: one read-only typed picture, every claim sourced and aged | 4 |
| S57 · Defragger & Context Compaction | absent — memory caps and retention sweeps exist, nothing compacts the working set | 4 |
| S58 · E.D.I.T.H. / Hasan-Jalal - one authorization plane for the fleet | built 2026-10-17 - one authorization plane: 12 action categories x 8 actors resolve to exactly one verdict each, an unknown action escalates rather than passing as harmless, a dangerous payload shape is refused for the owner too, every decision appends to the existing hash-chained trail with the actor that asked, and the gate going down reads rather than eats | 8 |
| S59 · FRIDAY / Artaxias - protocol authorization | built 2026-10-17 - every protocol resolves by index, codename or name; a spoken number is understood; failures back off and survive a restart but a correct password always works, so the owner cannot be locked out of his own machine; a one-time code is spent once. Second factor is optional and unset, so today the gate is password-only exactly as it was | 7 |
| S60 · U / Bagadata - the fleet's development agent | built and REACHABLE 2026-10-17 - convene() has a caller (brain/u_launch), seats get their own git worktree on a run-scoped branch, teardown refuses over uncommitted work, and a live run spawned an installed CLI with the owner's checkout untouched. 16 of 19 seats drivable. Not yet proven: a real seat answering, which needs a provider credential the allowlist withholds on purpose | 7 |
| S61 · DUM-E / Hugh - physical operations and house custody | built 2026-10-17 - the interlock is first and pure: an E-Stop reachable from every state, ceilings enforced AT THE INTERLOCK even against a valid authorisation, process identity protected inside the kill branch, and no software path that can clear a trip. The house is read as a fifth source in the existing sitrep through the one existing HA client. Printer, CUPS, glasses and workshop thermals remain deferred for want of the hardware | 12 |
Source: docs/systems-plan.json. Generated using scripts/build_website_inventory.py.