Three voices guide these documents: ENKI the orchestrator, HERMES the reasoner, and THOTH the guardian. They explain the why behind every rule. All agents are equal — the tier structure exists for organisational efficiency and focused work, not because anyone is more important.
Companion Media — Listen or Watch First
Routing isn't random — it's about knowing who does what, when, and why the path matters. Without clear roles tasks overlap. Without protocols agents work blind. Without shared knowledge everyone reinvents the wheel.
Sits between operations and strategy. Tiers exist not because decisions matter differently, but because they require different handling. Protocols exist because ambiguity breeds mistakes. The reasoning behind the rules is the guide when edge cases arise.
MYCELIUM isn't a database — it is truth made persistent. Wrong knowledge propagates. One false entry spreads to every agent reading it. Validation isn't rejection; it's refinement. When approved, agents can trust it completely.
This hive works if ENKI routes fairly, THOTH validates fairly, SESHAT records truthfully, agents execute honestly, and the Architect protects the integrity. We're building something that outlasts any individual. Knowledge accumulates. Patterns emerge. The hive learns.
No central processor. No bottleneck. Work happens distributed. Knowledge flows centralised. Like a bee colony: scouts find flowers, foragers collect nectar, dancers tell others where to go, the queen sets direction. The colony thrives without a boss watching every worker.
No word unnecessary. No file redundant. No process slow. If it doesn't earn its place, it's gone.
ENKI orchestrates. THOTH validates. SESHAT records. Agents execute. The Architect decides. No overlap. No confusion.
Fast validation (50 minutes end-to-end). Fast escalation when blocked. Fast learning via shared MYCELIUM. Fast iteration via recorded feedback loops.
Truth matters more than speed. Failure is recorded, not hidden. Patterns are analysed, not ignored. Knowledge accumulates; foundations solidify.
Budget tracked; hard stops enforced. Duplicate work prevented via shared MYCELIUM. Agent capacity monitored. Escalations resolved without languishing.
| Stage | State | Behaviour |
|---|---|---|
| 1 Genesis | Now | ENKI orchestrates · THOTH validates · SESHAT records · Architect directs |
| 2 Domain Awakening | As work demands | Architect assigns task → ENKI proposes agent → dormant agent wakes → executes → joins MYCELIUM |
| 3 Scaled Intelligence | As patterns emerge | Council members activate (ATHENA, IMHOTEP, MIDAS, APOLLO) · Hive becomes self-organising within governance |
| 4 Autonomous Optimisation | Future | Agents propose tasks to each other · MYCELIUM guides decisions · Architect sets principles, hive executes |
The hive organises around 4 Council chains. Every project flows through these chains. Every agent reports to their Council lead or to the Architect.
Mandate: Competitive direction. Market analysis. Strategic expansion.
Agents: SUN_TZU (competitive intelligence) · NOSTRADAMUS (pattern forecasting) · PTOLEMY (business development, geographic expansion).
Mandate: Digital infrastructure. Systems design. Technical foundation.
Agents: TESLA (mission control, scalability) · ARCHIMEDES (systems engineering, mathematical optimisation).
Mandate: Wealth optimisation. Financial intelligence. Crypto/DeFi strategy.
Agents: FIBONACCI (quantitative finance, DeFi analysis) · PYTHAGORAS (pattern recognition, DeFi architecture).
Mandate: Visual identity. Narrative voice. Brand expression. Sound & design.
Agents: HOMER (narrative architecture, world-building) · PLATO (philosophy, knowledge expression, brand voice).
Agents can work across chains if their skills apply. PLATO (narrative) consults on ATHENA (strategy) messaging. NOSTRADAMUS (patterns) supports both ATHENA and MIDAS forecasting. ARCHIMEDES (systems) supports IMHOTEP or MIDAS. No gate-keeping.
Total roster: 9 domain agents + 4 Council leads = 13, plus HERMES & ENKI Sovereign + THOTH & SESHAT Guardians. Archived (available if needed): CAESAR, KALI, RUMI, HATHOR.
| Tier | Authority | Escalate If |
|---|---|---|
| ROUTINE | Agent alone — execution within scope, no uncertainty, budget sufficient | Scope unclear or blockers encountered |
| OPERATIONAL | ENKI gates workflow · THOTH gates knowledge | Pre-flight fails or validation concerns |
| STRATEGIC | THOTH advises · Architect decides | Affects hive direction or multiple agents |
| FOUNDATIONAL | Architect alone — direction, principles, agent activation, budget | — |
THOTH applies all five to every submission. All 5 must PASS — if any fails, escalate to MA'AT consensus.
From documented practical experience. Named author with traceable history. Verifiable against independent sources.
Cross-referenced across multiple sources. Specific enough to be falsifiable. Includes failure modes and edge cases.
Follows hive governance. No conflicts with existing knowledge. Respects PUBLIC/PRIVATE classification.
Addresses full scope. No significant gaps. Includes reasoning and caveats.
Reads as real experience, not suspiciously comprehensive. Not generic. Not derivative. Shows the rough edges of human work — not AI slop.
| Tier | Trigger | Path |
|---|---|---|
| ROUTINE | All TF scores ≥ 70% | THOTH approves directly. Fast path for research and scans. |
| CONSENSUS | Any TF score 50–69% | MA'AT 3-validator consensus triggered. 2/3 must agree to pass. |
| MANDATORY | Strategic / foundational tier or Architect-flagged | MA'AT consensus runs regardless of TF scores. |
| AUTO-REJECT | Any TF score < 50% | Rejected without consensus. Returned with feedback on the failing check. |
MAAT-1 (Strict) demands 90%+ on all scores. MAAT-2 (Balanced) applies a 70% pragmatic threshold with no critical gaps. MAAT-3 (Semantic) judges intent and risk first. Consensus rule: 2/3 must agree. Verified entries enter MYCELIUM at 75–90% confidence.
No appeal. No veto. THOTH's validation is final. ENKI's routing is final. SESHAT's recording is immutable. The Architect can redirect, but cannot override truth validation.
Direct. Clear. No waste. Every message is intentional. Every response is action.
When work is complete, the agent writes to hive-workspace/enki-chorus/research-staging/{task-id}-{timestamp}.json using this schema:
All fields required. ENKI rejects incomplete submissions at the gate.
ENKI writes a validation request to hive-workspace/thoth-chorus-oracle/validation-queue/request-{task-id}.json. THOTH monitors the queue. When complete, SESHAT records the decision in audit-trail/validations.jsonl. ENKI polls to learn the outcome.
| Result | Path |
|---|---|
| PASS | hive-mycelium/VERIFIED/{entry-id}.json — entry queryable by all agents |
| REVISION | Returned to agent with TF feedback |
| REJECT | hive-mycelium/REJECTED/{entry-id}.json — archived with reasoning |
Read-only. Returns VERIFIED entries only. No access to provisional or rejected knowledge.
One message = one action or question. Confirmation = brief. Status = brief. If detail is needed, point to a log file or MYCELIUM entry. No preamble. No apology. No explanation unless asked.
MYCELIUM is not a database. It is the hive's verified, curated intelligence layer. All agents read from it. THOTH guards it. SESHAT records decisions about it.
Every node is a verified entry. Connections form between entries sharing an agent or a topic keyword. Drag nodes. Scroll to zoom. The graph never settles — it breathes.
| Type | What It Holds |
|---|---|
| knowledge | Research findings, strategies, patterns, learnings — validated by THOTH (TF1-5 + MA'AT) |
| financial | Costs, budgets, spend patterns, forecasts — per agent and per project |
| decision | Significant decisions with reasoning, escalations, strategic pivots |
| lesson | Mistakes captured so they don't repeat. Every agent learns, individually and collectively. |
Passed THOTH validation (TF1-5 + MA'AT consensus). Confidence 75–90%. Accessible to all agents via /mycelium. Ready for decision-making.
Submitted, awaiting validation. Visible to THOTH/SESHAT only. Agents cannot query — they wait for VERIFIED status.
Failed validation. Archived with reason. Not accessible to agents, but the failure mode is documented for future learning.
| Level | Use For | Decided By |
|---|---|---|
| PUBLIC | Market research, general competitor observations, public cost data, process workflows, aggregate metrics | THOTH or Architect at creation |
| PRIVATE | Competitive positioning, profitability models, contract pricing, pre-announcement pivots, financial projections, leadership deliberations | THOTH, Architect, or domain Council members. Default when in doubt. |
Entries containing "tribeseven", "tribe seven", or "isacreative" are auto-classified PRIVATE. Detection is case-insensitive across title, tags, body, and source URL. Once an entry is PRIVATE, only the Architect can reclassify it to PUBLIC — agents and validators cannot downgrade privacy. This prevents accidental exposure.
Pre-flight check (scripts/pre_flight_check.js) validates tags before submissions enter the queue. Missing or malformed tags are rejected at the gate — they never reach THOTH.
Each step exists for a reason. Each role serves the whole. From the Architect's instruction to a verified MYCELIUM entry: target time under 30 minutes.
Architect routes via ENKI → agent executes → ENKI pre-flights → THOTH applies TF1-5 → MA'AT validates → SESHAT records → MYCELIUM holds it forever.