This document provides a complete index of all documentation in the mindX system, organized by category and component type.
Last Updated: 2026-04-27 Total Documentation Files: 194+ CORE Documentation: Complete 15-component analysis ✅ Agent Documentation: 34+ Tool Documentation: 29+ DAIO Documentation: 15 files covering 12 core + modular contracts ✅ Production Ready: ✅ Enterprise-grade deployment and security documentation complete
prompts/boardroom/{role}.prompt (canonical character voice), agents/boardroom/{role}.agent (operating contract), agents/boardroom/{role}.persona (cognitive style as JSON). Boardroom._load_member_card() composes the system prompt with .prompt as primary, .persona traits/beliefs/priorities as enrichment, and .agent BOARDROOM ROLE as situational backup. _query_soldier(), roll_call(), and convene_stream() all inject the composed prompt. Endpoint GET /insight/boardroom/cards shows the exact prompt loaded plus sources_loaded (which of the three files contributed). Five LLM knobs (MAX_CONCURRENT=8, NUM_CTX=8192, NUM_PREDICT=2000, TEMPERATURE=0.3, ROLLCALL_NUM_PREDICT=120) are env-overridable and visible at /insight/boardroom/roles. The split concerns let prompts iterate per-day, personas per-week, contracts per-quarter without cross-contamination.BOARDROOM_INFERENCE_BACKEND={auto|vllm|ollama} routes the 8 boardroom members through a self-hosted vLLM server when one is reachable. vLLM continuous batching = all 8 prompts in flight on one model, generation interleaves at the per-token level, all finish ~together (vs sequential per-vote with Ollama). Three deploy paths documented: Modal/RunPod cloud GPU (~$0.50/hr), LAN GPU server, or vLLM CPU on the VPS (slower but free). auto mode lets you defer the deploy — when vLLM appears at BOARDROOM_VLLM_BASE_URL, the boardroom switches to it on next call without a restart. Health endpoint /insight/boardroom/health shows live reachability.agent_workspaces/{agent}/process_trace.jsonl is written by log_process and read by nobody — 19.5 GB orphan in one mastermind file; (2) IPFS offload Phase 8 dormant because no Lighthouse keys vaulted; (3) wisdom-tier _training.jsonl files write-only, no consumer. Six phases shipped this session: per-tier policy doc (memory_tiers.md), GET /insight/memory/audit (live byte counts), GET /insight/memory/tiers (policy + state), tier-aware memory_agent.prune_stm (14 d ↔ IPFS eligibility), agents/storage/agent_workspace_pruner.py (rotate-archive-delete with disk-headroom guard), POST /insight/memory/prune (operator-triggered). Live recovery freed 19.5 GB; appendix in the audit doc preserves the incident and the safety guards added (commit b6d943a5)./data path live." Eight tiers: information_stm (14 d → archive → IPFS → delete), knowledge_ltm / concept_dreams / wisdom_training (forever — substrates for THOT), dream_reports (90 d → archive), archive (90 d → IPFS → delete), thot_anchored (forever, immutable on chain), orphan_workspaces (rotate at 100 MB · archive at 30 d · delete at 90 d). Three hard guarantees: nothing under ltm/ ever pruned; no delete until IPFS CID confirmed; default dry_run=True everywhere. Live state at /insight/memory/tiers.agents/cognition/wisdom_loader.py (NEW · 2026-04-27) — closes the cognitive-ascent loop's ingestion edge. Reads *_training.jsonl rows produced by every dream cycle, embeds the assistant message + first 200c of context into pgvector via the existing memory_pgvector.embed_and_store_doc() with a wisdom:{agent}:{ts}:{idx} doc-name prefix. Idempotent on re-index. Three endpoints: POST /insight/cognition/wisdom/index?hours=24 (bulk re-index recent files), GET /insight/cognition/wisdom/search?q=&top_k=3 (semantic retrieval — designed for BDI perceive() consumption), GET /insight/cognition/wisdom/stats (DISTINCT count). Auto-fires from machine_dreaming.run_dream_cycle after _write_training_data; first run indexed 91 wisdom docs on production. Phase 6b — BDI plan() prompt prepends "RELEVANT WISDOM" preamble — is the next step that closes the loop end-to-end.Boardroom._diagnose_recovery() pattern-matches the storm and emits a structured recovery block on every rollcall response. Four patterns wired today: ollama_signin_lapsed (auto-fires /insight/boardroom/cloud_signin, severity high), cold_load_storm (3 remediation hints, no auto-action), empty_ack_drift (diagnoses prompt/num_ctx truncation), persona_files_missing (names the affected seats). Surfaced as a banner above /feedback.html#sec-board rollcall + as CEO → SELF dialogue messages on /boardroom. Adding a 5th pattern is one entry — matcher predicate + recovery dict. Foundation for mindX adapting to its own errors./dreams.html — machine.dreaming with diagnostics — dedicated page for the 7-phase consolidation cycle. Hero strip shows total dreams · STM bytes freed · LTM bytes grown · median compression ratio · training rows · last cycle age. Operator-triggered "Accelerate dream now" button POSTs /insight/dreams/run for an immediate cycle. Recent-cycles table with byte-delta colour coding (green = STM freed, cyan = LTM grown, amber = compression ratio). Click any cycle → /insight/dreams/diff/{filename} drill-down with per-agent cards: STM input excerpts (raw memory files), LTM consolidated JSON, and the first training-jsonl rows. Each dream now writes a parallel *_training.jsonl in data/memory/ltm/<agent>/ — finetuning-ready chat-completion examples ({messages: [{role: system|user|assistant, ...}]}) suitable for OpenAI/Anthropic API or local LoRA. Substrate for promoting verified concepts to machine.wisdom and on-chain THOT mints.mindx.pythai.net/boardroom = CEO + 7 soldiers (CISO/CRO veto-weight 1.2×, CFO/CLO/CPO/CTO/COO weight 1.0×; supermajority 0.666). General-purpose AI / agent / member interaction: any participant whose identity verifies by signature can hold a seat. Decisions emerge from consensus + signature verification; the engine works with or without DAIO control (DAIO is the optional Solidity controller — when wired, decisions become on-chain enforced; when not, they remain off-chain). DAIO voting bridge: a boardroom-approved CEO decision is castable as the AI vote inside DAIO's 2/3 consensus across the Marketing / Community / Development groups (each group: 2 humans + 1 AI; the AI vote = boardroom output). One off-chain consensus → one on-chain ballot. Running in daio/governance/boardroom.py. Surfaced on /feedback.html#sec-board.
(2) Dojo — dispute resolution opened by any boardroom seat-holder when consensus fails or a decision needs escalation. Configurable consensus models per dispute type — 2/3, 50/99, supermajority, etc. — chosen at convocation.
(3) War council — foreign entity at mastermind.pythai.net. Isolated from mindX; consumes the mindX API as a paying external client. Monetized via bankon.pythai.net — every API call from the war council settles through BANKON, turning agent/inference/identity provision into mindX revenue. 13-seat on-chain assembly (CEO + GENIUS / SENATUS / CENSURA / FIDES / TABULARIUM / TESSERA / SPONSIO PACTUM / GUARDIAN / STRATEGIC EVOLUTION / COORDINATOR / TREASURY STEWARD / DOJO MASTER) tied to BONAFIDE contracts on Polygon mainnet. PYTHAI naming is a toggle within the war council. mindX provides agents, inference, and identity; the war council provides on-chain finality; BANKON provides the meter.
Variations on both boardroom and dojo are supported; CEO + 7 soldiers stays the primary decision tier.
agents/catalogue/ writing to data/logs/catalogue_events.jsonl./feedback.html — Live mind-of-mindX page. Agent dialogue (SSE), improvement ledger ⨝ gödel rationale, boardroom decisions with vote chips, dream cycles table, stuck-loop detector, inter-agent SVG ring graph, memories on chain section, inference health./feedback.txt — Public plain-text snapshot for watch curl …. ~24 lines covering storage, dreams, loops, last-10 dialogue.?h=true plain-text mode on every /insight/ and /storage/ endpoint — text/plain rendering via mindx_backend_service/text_render.py. JSON unchanged when omitted.agents/storage/ — IPFS via Lighthouse + nft.storage with deterministic byte-stable CAR bundles, sha256 verification, ARC DatasetRegistry chain anchor. Phase 8 of machine_dreaming.run_full_dream() triggers automatically when IPFS keys are vaulted.mindx_backend_service/main_service.py:_periodic_dream_cycle wrapped in resilient try/except; warmup 600s → 60s. Was silently dead since 2026-04-12.agents/core/bdi_agent.py now ships skeleton fallback that picks from registered tools (was hardcoding unregistered audit_and_improve); agents/monitoring/token_calculator_tool.py indentation bug fixed (UnboundLocalError on every plan call).data/config/{llm_factory_config,mindx_config}.json had a duplicate "llm" key + Mistral as default + Ollama disabled. Reordered to ollama-first; default model qwen3:1.7b not the not-pulled mistral-nemo:latest.mindX provides an API for agents, UIs, and external systems. When the backend is running (default port 8000):
AgenticPlace uses mindX as its provider; see AgenticPlace_Deep_Dive.md and API.md.
Whether mindX is ingesting, providing inference, or services, monitoring and rate control are essential in both directions (inbound and outbound). Actual network and data metrics are in scientific units (ms, bytes, req/min).
GET /api/monitoring/inbound — inbound request metrics (latency_ms, request/response bytes, requests_per_minute, rate_limit_rejects)llm/rate_limiter.py, api/ollama/ollama_url.py and other provider APIs expose get_metrics(); see rate_limiting_optimization.mdmindX is enhanced with DAIO (Decentralized Autonomous Intelligence Organization) - a comprehensive blockchain-native governance and economic layer that enables autonomous agent coordination with cryptographic sovereignty.
daio/governance/boardroom.py. Dojo (escalation): boardroom seat-holders open disputes; configurable consensus (2/3, 50/99, supermajority). War council (foreign entity, roadmap) at mastermind.pythai.net: 13 seats on BONAFIDE mainnet with PYTHAI naming as a toggle. Isolated from mindX — consumes the mindX API as a paying client; metered through bankon.pythai.net.Key DAIO Integration Points:
All agent documentation is organized in the agents/ folder structure. See agents/index.md for the complete agent registry.
Primary CORE References:
CORE Component Documentation:
Meta-Orchestration:
agents/core/mindXagent.py) - Meta-orchestrator with complete system understanding (~149KB)Cognitive Architecture:
Infrastructure Services:
Orchestration Layer:
Total CORE: ~582KB of foundational code enabling all system functionality
simple_coder_sandbox/update_requests.jsonComplete Agent Registry: See agents/index.md for full details, NFT metadata, and registry information.
All tool documentation is organized in the tools/ folder. Tools are now organized into subfolders for better structure and expansion. See TOOLS_INDEX.md for the complete tool registry and tools_organization_audit.md for the organization audit.
tools/core/)tools/financial/) - CFO Priority Accesstools/registry/)tools/communication/)agent.persona defines behavioral traits, agent.prompt defines specific instructions.tools/monitoring/)tools/development/)tools/identity/)tools/ root)api/ollama/ollama_model_capability_tool.pyapi/ollama/ollama_chat_display_tool.pyComplete Tool Registry: See TOOLS_INDEX.md for full details and usage information.
Tools Organization: See tools_organization_audit.md for comprehensive audit, organization structure, CFO priority access implementation, and production readiness checklist.
vault_manager.js. Optional access gate (NFT/fungible) for session issuance; see LIT_AND_ACCESS_ISSUANCE. DAIO keyminter contracts (VaultKeyDynamic, VaultKeyIntelligent) mint vault access keys.Complete blockchain-native governance and economic layer for autonomous agent coordination
Primary DAIO Documentation:
DAIO Core Contracts (12 Foundational + Modular Extensions):
Core Governance Contracts (4):
Identity & Agent Management (3):
Economic Infrastructure (3):
Voting & Extensions (2):
13th+ Contract - Consensys-Driven Service Branching:
DAIO Integration Guides:
ARC Protocol Integration (5 contracts):
Key DAIO Features:
./deploy/production_deploy.shInteractive API reference: When the backend is running (default port 8000), http://localhost:8000/docs (FastAPI Swagger UI) shows all API endpoints and lets you try requests and inspect schemas—the best way to explore and audit API interactions.
External Intelligence Agnostic Architecture: mindX is designed to be external intelligence agnostic. The system does not depend on any specific external LLM provider until mindXagent replicates itself into a model. During this replication process, mindXagent makes decisions from a choice of THOT (Transferable Hyper-Optimized Tensors) processes, enabling the creation of sovereign AI models.
Current Implementation:
THOT-Based Model Replication: When mindXagent replicates itself into a model, it uses the THOT ecosystem to create, optimize, and deploy neural network tensors as tradeable NFT assets. This enables mindX to evolve from external intelligence dependency to sovereign AI model creation.
The THOT ecosystem provides comprehensive lifecycle management for neural network tensors as tradeable NFT assets. This system enables mindXagent to replicate itself into sovereign AI models, making decisions from a choice of THOT processes rather than depending on external intelligence providers.
Key Concept: mindX is external intelligence agnostic until mindXagent replicates itself into a model. During replication, mindXagent selects from available THOT processes to create optimized neural network models that can operate independently.
Core Contracts (daio/contracts/THOT/core/):
Lifecycle Management (daio/contracts/THOT/lifecycle/):
Marketplace (daio/contracts/THOT/marketplace/):
Integration Layer (daio/contracts/THOT/integration/):
NFT Components (daio/contracts/THOT/nft/):
Automation (daio/contracts/THOT/agents/):
Interfaces (daio/contracts/THOT/interfaces/):
Architecture Documentation:
User Journeys:
agents/utility/)GET /users/session/validate, POST /users/logout; vault user folders GET/PUT/DELETE /vault/user/keys; access gate (optional ERC20/ERC721); frontend vault_manager.js; LIT_AND_ACCESS_ISSUANCE; DAIO keyminter (VaultKeyDynamic, VaultKeyIntelligent)All agent and tool documentation follows consistent standards:
See agents/index.md for agent documentation standards and TOOLS_INDEX.md for tool documentation standards.
mindX is external intelligence agnostic until mindXagent replicates itself into a model. The system architecture is designed to:
External Intelligence Agnostic Phase
↓
Ollama GPU Inference (10.0.0.155:18080)
↓
mindXagent Self-Replication
↓
THOT Process Selection
↓
Sovereign AI Model Creation
See System Architecture Map for complete architectural details.
mindx_backend_service/bankon_vault/vault.py, credential_provider.py, routes.pymanage_credentials.py — store, list, delete, providersconfig/providers/.envagents/memory_pgvector.py — async connection pool, dual-write with file fallbackscripts/migrate_to_pgvector.pyllm/inference_discovery.py — probes all sources, auto-selects best providerllm/multi_stream.py — parallel queries with consensus strategiesllm/vllm_handler.py — OpenAI-compatible API integration/inference/status, /inference/probe, /inference/multi-streamdaio/contracts/algorand/bona_fide.algo.ts — Algorand verification token with clawbackdaio/agents/agent_map.json — canonical registry with reputation, groups, verification tiersdaio/governance/boardroom.py — CEO + Seven Soldiers weighted consensusdaio/governance/dojo.py — reputation ranks (Novice→Sovereign), BONA FIDE privilege system/boardroom/convene, /dojo/standings, /dojo/agent/{id}mindx_backend_service/dashboard.html — animated agent network, real-time metrics/ (dashboard), /diagnostics/live (JSON), /docs.html (documentation hub)agents/author_agent.py — publishes The Book of mindX every 2 hoursagents/learning/improvement_journal.py — auto-updates every 30 min/doc/{name} — renders any of 194 docs as styled HTML/book, /journal, /docs.html (TOC)X-Session-Token or Authorization: Bearer <api_key>/, /health, /docs.html, /doc/, /book, /journal, /diagnostics/live, /dojo/standingsscripts/genesis_production_identities.py — generates fresh wallets, stores in BANKON Vaultdata/identity/production_registry.json (public addresses only)Last Updated: 2026-04-03 (Documentation Audit & pgvector Migration) Maintained By: mindX Documentation System + AuthorAgent Production Status: ✅ Enterprise-ready with pgvector, BANKON Vault, live diagnostics For Issues: See project repository