POWERFULMOVES/PMOVES.AI

PMOVES.AI — Decentralized multi-agent AI platform by Cataclysm Studios Inc. Orchestrators, services, knowledge bases, and security hardening.

7

stars

3,951

commits

Python

primary language

Sep 9, 2026

updated

cataclysmstudios.net

README

PMOVES.AI

Integration Contract CodeQL Advanced CHIT Contract Check Docker Hardening Validation Python Tests

A local-first, multi-agent orchestration platform built on a rooms-on-a-stage model. P7 (Pinokio 7) is the room-aware stage manager that selects rooms, loads suits, and manages stage transitions (rehearsal, live, review, archive). Each room is an agent-owned workspace with bound services, skills, and notebook state — wired together with NATS event-driven messaging and full Prometheus/Grafana/Loki observability.

PMOVES is structured as a Metal-Organic Framework (MOF) for distributed machine intelligence — the crystalline lattice through which autonomous agents flow. Operationally, this translates to a rooms-on-a-stage model: P7 (Pinokio 7) is the evolving room-aware stage manager that selects rooms, loads suits, and manages stage transitions (rehearsal → live → review → archive). This model grows with the platform — new rooms, stage states, and suit types are added as the topology evolves.

Quick Start

make first-run

This single command orchestrates the full onboarding sequence: environment prompts, Supabase CLI bring-up, data/service seeding, core + agent + external stacks, and the 12-step smoke harness. When it finishes successfully every bundled integration (Wger, Firefly, Jellyfin, Open Notebook, Agent mesh) is online with branded defaults. See the First-Run Bootstrap Overview for a detailed breakdown of each step.

Key Directories

  • CATACLYSM_STUDIOS_INC/ – Provisioning bundles and infrastructure automations for homelab and field hardware, including unattended OS installs, Jetson bootstrap scripts, and ready-to-run Docker stacks that mirror the production mesh topology.
  • docs/ – High-level strategy, architecture, and integration guides for the overall PMOVES ecosystem. See also pmoves/docs/ENVIRONMENT_POLICY.md for the single-file environment policy and Jellyfin host-mount instructions.
  • pmoves/ – The primary application stack with docker-compose definitions, service code, datasets, Supabase schema, and in-depth runbooks for daily operations and advanced workflows.
  • pmoves/contracts/solidity/ – Hardhat workspace prototyping Food-USD / GroToken governance flows with automated tests that model staking, quadratic voting, and group-buy execution.
  • pmoves/ui/ – Next.js + Supabase Platform Kit workspace for the upcoming web UI; reuses pmoves/.env.local so frontend hooks can target the same Supabase CLI stack.

Essential Documentation

  • Claude Code CLI Integration – TAC integration with custom slash commands, security hooks, and PMOVES-aware context for AI-assisted development.
  • Testing Strategy – Comprehensive testing guide covering smoke tests, functional tests, and end-to-end validation workflows.
  • PMOVES Stack README – Quickstart environment setup, service inventory, and Codex bootstrap steps for running the orchestration mesh locally.
  • Local Tooling Reference – One-stop index for environment scripts, Make targets, Supabase workflows, smoke tests, and provisioning helpers.
  • Supabase Service Guide – CLI vs compose expectations, realtime wiring (supabase start --network-id pmoves-net), and how PMOVES consumes PostgREST/Realtime in both local and self-hosted deployments.
  • PMOVES Docs Index – Curated entry points into the pmoves-specific runbooks covering Creator Pipeline, ComfyUI flows, reranker configurations, and smoke tests.
  • UI workspace bring-up – Next.js + Supabase quickstart (npm/yarn commands, env loading from pmoves/.env.local, Supabase CLI prerequisites).
  • Service Docs Index – Per-service guides (overview, compose/ports, runbooks, smoke tests, and roadmap alignment).
  • External Integrations Bring-Up – Wger, Firefly III, Open Notebook, and Jellyfin commands, token wiring, and port overrides for make up-external.
  • Architecture Primer – Deep dive into mesh topology, service responsibilities, and evolution of the orchestration layers.
  • Complete Architecture Map – Full-fidelity view of the latest integration mesh, including data planes and edge deployments.
  • Multi-Agent Integration Guidelines – Operational patterns for coordinating Agent Zero, Archon, and automation hubs across environments.
  • Archon Updates for PMOVES – What changed in the October 2025 Archon bundle, how to wire the Supabase CLI stack, and the MCP/NATS expectations.
  • Make Targets Reference – Command catalog for starting, stopping, and tailoring compose profiles (core data plane, media analyzers, Supabase modes, and agent bundles).
  • Single-User (Owner) Mode – Personal-first operation without login prompts; boot-JWT auto-auth, owner chip in the UI, and security notes.
  • Production Readiness Report – Feb 2026 audit of service health, security posture, and deployment readiness.
  • Codex + Copilot Review Workflow – How to combine the Codex CLI reviewer with GitHub Copilot's PR assistant, including token setup and evidence logging expectations.

CHIT & Geometry Documentation

Compressed Hierarchical Information Transfer (CHIT) and the Geometry Bus are core to how PMOVES.AI encodes, routes, and decodes structured knowledge across the agent mesh.

See pmoves/docs/PMOVESCHIT/ for the full 19-document collection including decoder specifications, neural-network notebooks, audit reports, and UI design specs.

Rooms on a Stage

PMOVES' Metal-Organic Framework architecture manifests operationally as rooms on a stage — a living topology (not a frozen specification) with three layers:

  • Rooms are the agent-owned entrypoints: infra fabric, field control, voice studio, workstation. Each room binds services, skills, and notebook state.
  • Stage is the lifecycle state per room: rehearsallivereviewarchive. Rooms transition through these states as work progresses.
  • Suits are the runtime/persona bindings layered onto rooms — upstream Agent Zero as the external baseline, PMOVES hardened overlays as the custom fit, voice/theme/persona as the visible styling layer.

P7 (Pinokio 7) is the evolving room-aware stage manager: it reads the room catalog, selects the appropriate room profile for a given workload, loads the correct suit, and manages stage transitions. P7's NATS subjects (p7.nats.launch, p7.nats.session) are the control plane for room entry and lifecycle. As the platform grows, P7's tree adds new rooms, stage states, and suit types — the topology is designed to expand.

Rooms own presentation and session ergonomics; the notebook plane owns durable memory. The Room Manifest Contract defines the interface; pmoves/config/rooms/catalog.json is the canonical seed catalog. See Rooms on a Stage Overview for the end-to-end model and AGNOTE4482 for the full P7 specification.

Room Index

🏗️ Z890 Infra Fabric Room

Purpose: Infrastructure room for topology, service health, secrets discipline, and operator bring-up. Profile: z890-infra · Agent: z890-claude · Stage: live · Manifest: z890-infra.room.fabric.json

  • pmoves/services/node-registry/ — Multi-host node discovery and health reporting
  • pmoves/services/resource-detector/ — Hardware capability detection (GPU, CPU, memory)
  • pmoves/services/work-marshaling/ — Distributed task scheduling and work queue management
  • pmoves/services/gpu-orchestrator/ — GPU resource allocation and scheduling
  • pmoves/services/vllm-orchestrator/ — vLLM inference server lifecycle management
  • pmoves/services/model-registry/ — Central model catalog with version tracking
  • pmoves/services/tensorzero-config-api/ — Dynamic TensorZero model configuration API
  • pmoves/services/benchmark-runner/ — Automated performance benchmarking harness
  • pmoves/services/nats-echo/ — NATS message debugging and replay tool
  • pmoves/services/analysis-echo/ — Analysis pipeline event echo and auditing
  • pmoves/services/evoswarm/ — Evolutionary swarm optimization coordinator

🔭 4090 Field Control Room

Purpose: Scout/control room for review, topology, handoff, and notebook-backed triage. Profile: 4090-field · Agent: 4090-claude · Stage: live · Manifest: 4090-field.room.control.json

  • pmoves/services/agent-zero/ — MCP bridge + decision engine (ingests Supabase + CHIT events)
  • pmoves/services/archon/ — Agent builder/knowledge management with Supabase CLI realtime + NATS clients
  • pmoves/services/deepresearch/ — Tongyi DeepResearch bridge with OpenRouter/local modes plus Open Notebook mirroring
  • pmoves/services/supaserch/ — Multimodal holographic deep research orchestrator
  • pmoves/services/graph-linker/ — Knowledge graph linking and entity relationship management
  • pmoves/services/session-context-worker/ — Session context aggregation for multi-turn agent conversations
  • pmoves/services/gateway-agent/ — Unified API gateway with agent-aware routing
  • pmoves/services/botz-gateway/ — Skills marketplace gateway for BoTZ agent capabilities
  • pmoves/services/agentgym-rl-coordinator/ — Reinforcement learning coordinator for agent skill training
  • pmoves/services/consciousness-service/ — Agent self-model and meta-cognitive state tracking
  • pmoves/services/chat-relay/ — Agent-to-agent and agent-to-user chat relay
  • pmoves/services/messaging-gateway/ — Multi-channel messaging gateway
  • pmoves/services/a2ui-nats-bridge/ — Agent Zero UI to NATS event bridge
  • pmoves/services/open-notebook/ — Streamlit UI + SurrealDB API (container ports 8502/5055 per upstream; host defaults map to :8503 UI and :5055 API, override with OPEN_NOTEBOOK_*_PORT) for research assets and MCP notebooks. Doc: pmoves/docs/services/open-notebook/
  • pmoves/services/notebook-sync/ — Bridges Open Notebook datasets into Supabase and LangExtract flows
  • pmoves/services/retrieval-eval/ — Retrieval benchmarking, relies on Supabase + hi-rag

🎙️ 5090 Voice Studio

Purpose: Voice-first room for TTS, media pipelines, notebook-backed iteration, and audition workflows. Profile: 5090-voice · Agent: 5090-claude · Stage: live · Manifest: 5090-voice.room.studio.json

  • pmoves/services/flute-gateway/ — Multimodal voice communication layer (HTTP :8055, WebSocket :8056) with Pipecat integration and prosodic synthesis
  • pmoves/services/vibevoice-realtime/ — Real-time voice synthesis service
  • pmoves/services/pmoves-yt/ — YouTube ingest; publishes geometry packets after segmentation
  • pmoves/services/channel-monitor/ — External content watcher; triggers ingestion on new uploads
  • pmoves/services/publisher/ — Discord & Jellyfin publisher with geometry-aware payloads
  • pmoves/services/publisher-discord/ — Dedicated Discord notification bot for ingest/summary events
  • pmoves/services/jellyfin-bridge/ + pmoves/docs/services/jellyfin-ai/ — Media sync bridging Jellyfin metadata into Supabase + Discord publisher
  • pmoves/services/{presign,render-webhook,extract-worker,langextract,media-audio,media-video,pdf-ingest,comfy-watcher,comfyui} — Supporting ingestion, extraction, and media tooling

💻 5090 KiloCode GLM Workstation

Purpose: GPU inference specialist node running KiloCode GLM on the GLM Coding Plan. Shares the 5090 with Claude Code and Codex. Profile: 5090-kilocode · Agent: 5090-kilocode · Stage: rehearsal · Manifest: 5090-kilocode.room.studio.json

  • pmoves/services/tensorzero-config-api/ — Dynamic TensorZero model configuration API (shared with Infra room)
  • pmoves/services/gpu-orchestrator/ — GPU resource allocation and scheduling (shared with Infra room)
  • pmoves/services/model-registry/ — Central model catalog with version tracking (shared with Infra room)
  • pmoves/services/retrieval-eval/ — Retrieval benchmarking (shared with Field Control room)

Cross-Cutting: Geometry & CHIT Fabric

These services form the knowledge and geometry substrate available to all rooms via the NATS event bus:

  • pmoves/services/hi-rag-gateway-v2/ — v2 gateway (CPU :8086, GPU :8087). Handles /geometry/*, jump, decode, calibration, Supabase realtime warmups, and CGP persistence.
  • pmoves/services/hi-rag-gateway/ — v1 legacy gateway (host :8089). Minimal CHIT endpoints for backward compatibility.
  • pmoves/services/gateway/ — Experimental CHIT UI/API for live geometry visualisation and WebRTC broadcast.
  • pmoves/services/mesh-agent/ — Geometry mesh bridge; signs and republishes geometry.cgp.v1 across deployments.
  • pmoves/services/evo-controller/ — Geometry tuning controller; reads CGPs from Supabase, emits tuning capsules back into the bus.
  • pmoves/services/tokenism-simulator/ — Token geometry simulation and visualization.

Cross-Cutting: Orchestration & External Data

  • pmoves/services/n8n/ — Workflow orchestrator; health/finance webhooks emit CGPs via hi-rag v2.
  • pmoves/services/wger/ — Health metrics ingest (paired with Supabase tables + health.weekly.summary.v1 CGPs).
  • pmoves/services/firefly-iii/ — Personal finance ingest; finance flows create finance.monthly.summary.v1 CGPs.

See each directory's README for ports, Make targets, and geometry notes. New integrations reference external repositories under integrations-workspace/ and the setup steps captured in pmoves/docs/EXTERNAL_INTEGRATIONS_BRINGUP.md.

Dashboards & UIs (local defaults)

By Room

Z890 Infra Fabric Room:

4090 Field Control Room:

5090 Voice Studio:

5090 KiloCode GLM Workstation:

Default access and operator credentials

  • Supabase operator is provisioned by make supabase-boot-user (also run by make first-run). The command writes values to pmoves/env.shared and pmoves/.env.local:

    • SUPABASE_BOOT_USER_EMAIL, SUPABASE_BOOT_USER_PASSWORD, SUPABASE_BOOT_USER_JWT.
    • The PMOVES UI auto-authenticates with NEXT_PUBLIC_SUPABASE_BOOT_USER_JWT so most routes won't prompt for a password. If you need to log in manually, use the email/password above from your env files.
  • Jellyfin uses the LinuxServer image defaults. After first boot, confirm the admin user and API key in pmoves/env.jellyfin-ai or via the Jellyfin UI (Settings → Dashboard). Update JELLYFIN_API_KEY and JELLYFIN_USER_ID in pmoves/env.shared if you rotate.

  • Wger and Firefly are brought up with PMOVES-branded defaults sourced from pmoves/env.shared (see pmoves/docs/FIRST_RUN.md "Seeded & Branded Defaults" for the exact initial usernames and emails).

  • Open Notebook's UI password also serves as its API bearer token; keep OPEN_NOTEBOOK_API_TOKEN identical to OPEN_NOTEBOOK_PASSWORD so CLI helpers and agents work against the same branded login (see pmoves/docs/services/open-notebook/README.md).

  • For a full list of seeded branded logins and where they come from, see:

    • pmoves/docs/FIRST_RUN.md (Seeded & Branded Defaults)
    • docs/SECRETS.md (Secret Management Playbook)
  • Creator bundle: see pmoves/creator/ for installers, tutorials, and ComfyUI workflows supporting WAN Animate, Qwen Image Edit+, and VibeVoice TTS. Key guides include:

  • Creator Pipeline Runbook – Current status of n8n automations (health/finance live, creative flows staging) plus geometry mapping and persona playback prep.

Developer Tools & Testing

Claude Code CLI Integration (TAC)

PMOVES.AI integrates deeply with Claude Code CLI through the .claude/ directory, providing context-aware development assistance:

Key Features:

  • Always-on context (.claude/CLAUDE.md) - Makes Claude aware of PMOVES architecture, services, and patterns
  • Custom slash commands - Quick access to production services:
    • /health:check-all - Verify all services are running
    • /search:hirag "query" - Query Hi-RAG v2 knowledge base
    • /agents:status - Check Agent Zero orchestration status
    • /deploy:smoke-test - Run integration test suite
  • Security hooks - Pre-tool validation blocks dangerous operations (rm -rf /, DROP DATABASE, etc.)
  • Observability hooks - Post-tool events published to NATS (claude.code.tool.executed.v1)

Quick Start:

# Use Claude Code CLI with PMOVES context
claude

# Try a custom command
/health:check-all

# Query knowledge base
/search:hirag "How does Agent Zero work?"

Learn more: .claude/README.md | Developer Context

Testing Infrastructure

PMOVES.AI employs comprehensive testing from smoke tests to end-to-end workflows:

1. Smoke Tests (30-60 seconds) - Quick health validation:

cd pmoves
make verify-all          # Full smoke test suite
make smoke               # Core services only
make smoke-gpu           # GPU-enabled services
make archon-ui-smoke    # Archon agent services
make deepresearch-smoke  # Research orchestration

2. Functional Tests (2-5 minutes) - Feature validation:

# Run all functional tests
cd pmoves/tests
./run-functional-tests.sh

# Individual test categories
./run-functional-tests.sh --category retrieval
./run-functional-tests.sh --category agents
./run-functional-tests.sh --category ingestion

3. Integration Tests - End-to-end workflows:

# NATS event flow validation
make test-nats-flow

# Media pipeline validation
make test-media-pipeline

# Agent coordination validation
make test-agent-mesh

Coverage Map:

  • Service health endpoints (Qdrant, Neo4j, Meilisearch, Supabase)
  • Retrieval accuracy (Hi-RAG v2 reranking, embedding generation)
  • Agent coordination (Agent Zero MCP, Archon prompts, NATS routing)
  • Media processing (YouTube ingestion, Whisper transcription, YOLO analysis)
  • Observability (Prometheus metrics, Grafana dashboards, Loki logs)

Learn more: docs/testing/TESTING.md

Development Workflow

Typical developer flow:

  1. Environment setup - make first-run or make bootstrap
  2. Service verification - make verify-all (smoke tests)
  3. Development - Use Claude Code CLI with custom commands
  4. Testing - ./run-functional-tests.sh before commits
  5. Monitoring - Grafana dashboard at http://localhost:3000

Best practices:

  • Use /health:check-all before starting work to verify service status
  • Leverage existing services (Hi-RAG, SupaSerch, Agent Zero) - don't duplicate
  • Publish events to NATS for async coordination
  • Follow observability patterns (expose /healthz and /metrics)
  • Run smoke tests after infrastructure changes
  • Use security hooks to prevent dangerous operations

Build Status & Security

Security posture: 29 open CodeQL alerts (2 critical, 22 high, 5 medium) — triaged into 7 remediation groups. PRs #651/#653/#654 resolved 36 alerts, but the Hardened branch scope re-surfaced 29 on a broader CodeQL scan. The 2 critical alerts are SSRF findings in Hi-RAG gateway services requiring URL allowlisting.

Pre-production blockers (6 remaining): AB-1 (A2UI nested gitlink), AB-3 (GHCR triggers), AB-4 (real credentials), AB-5/AB-6 (runtime validation, deferred until AB-4), AB-7 (PBKDF2 iteration bump). See the Production Audit Dashboard for full details and resolution sequence.

CI gates (all enforced on PRs to main):

  • CodeQL Advanced — Static analysis for JS/Python/Go vulnerabilities
  • CHIT Contract Check — Schema validation for geometry packet contracts
  • SQL Policy Lint — Migration and policy validation
  • Docker Hardening Validation — Container security baseline enforcement
  • Integration Contract — Cross-service API contract verification
  • Python Tests — Unit and integration test suite

Key security PRs:

  • #651 – Initial CodeQL alert triage (36 → 12)
  • #653 – Remaining CodeQL fixes (12 → 6)
  • #654 – Final 6 alerts resolved across gateway and YT services
  • Note: 29 alerts re-surfaced on the Hardened branch due to expanded CodeQL scan scope

Docker build reliability: All core services build successfully. See Build Fixes Documentation for historical context on DeepResearch, FFmpeg-Whisper, and environment file fixes.

Production readiness: See the Production Audit Dashboard for the consolidated audit covering active blockers, CodeQL triage, static audit layers, and resolution sequence. Historical context: Feb 2026 Readiness Report.

Getting Started

  1. Bootstrap the stack – For brand-new machines run make first-run. For incremental setup follow the environment and container launch instructions in the pmoves/README.md: place overrides in pmoves/.env.local, run make bootstrap to capture credentials, make up to start the core services, and make bootstrap-data to apply Supabase SQL, seed Neo4j, and load the demo Qdrant/Meili corpus before smoke testing.
  2. Review orchestration flows – Use the Make Targets Reference for day-to-day compose control, and consult the architecture and multi-agent guides in /docs for how Agent Zero, Archon, and supporting services communicate across the mesh.

Initial Setup & Tooling Flow (manual path)

  1. Environment bootstrap – Walk through pmoves/README.md to provision runtime prerequisites, seed pmoves/env.shared, and populate secrets. Use make bootstrap (wrapping python -m pmoves.scripts.bootstrap_env) when you need finer control, or invoke python3 -m pmoves.tools.mini_cli bootstrap --accept-defaults to script the same flow alongside the provisioning bundle. Both paths update env.shared, .env.generated, .env.local, and the auxiliary env.*.additions files consumed by Compose and the UI launcher.
  2. Supabase realtime alignment – Follow the Supabase Service Guide to start the CLI stack with supabase start --network-id pmoves-net (run this before accepting Supabase prompts in make bootstrap) and mirror the websocket endpoint (SUPABASE_REALTIME_URL=ws://host.docker.internal:65421/realtime/v1). This matches our self-hosted Supabase deployments.
  3. UI workspace bring-upcd pmoves/ui then npm install (or yarn install). The Next.js app loads Supabase creds from pmoves/.env.local and expects the Supabase CLI stack (make supa-start + make supa-status) before running npm run dev.
  4. Tooling cheatsheet – Keep Local Tooling Reference handy for Make targets, smoke tests, and environment scripts (env_setup, flight-check, smoke).
  5. Provisioning & hardware targets – Browse CATACLYSM_STUDIOS_INC/ for automated OS images, Jetson bootstrap bundles, and pmoves-net Docker stacks ready for edge hardware.

Contributors

POWERFULMOVES

3,095 commits

hunnibear

677 commits

dependabot[bot]

138 commits

Barathicite

15 commits

POWERFULMOVES/PMOVES.AI

PMOVES.AI — Decentralized multi-agent AI platform by Cataclysm Studios Inc. Orchestrators, services, knowledge bases, and security hardening.

7

stars

3,951

commits

Python

primary language

Sep 9, 2026

updated

cataclysmstudios.net

README

PMOVES.AI

Integration Contract CodeQL Advanced CHIT Contract Check Docker Hardening Validation Python Tests

A local-first, multi-agent orchestration platform built on a rooms-on-a-stage model. P7 (Pinokio 7) is the room-aware stage manager that selects rooms, loads suits, and manages stage transitions (rehearsal, live, review, archive). Each room is an agent-owned workspace with bound services, skills, and notebook state — wired together with NATS event-driven messaging and full Prometheus/Grafana/Loki observability.

PMOVES is structured as a Metal-Organic Framework (MOF) for distributed machine intelligence — the crystalline lattice through which autonomous agents flow. Operationally, this translates to a rooms-on-a-stage model: P7 (Pinokio 7) is the evolving room-aware stage manager that selects rooms, loads suits, and manages stage transitions (rehearsal → live → review → archive). This model grows with the platform — new rooms, stage states, and suit types are added as the topology evolves.

Quick Start

make first-run

This single command orchestrates the full onboarding sequence: environment prompts, Supabase CLI bring-up, data/service seeding, core + agent + external stacks, and the 12-step smoke harness. When it finishes successfully every bundled integration (Wger, Firefly, Jellyfin, Open Notebook, Agent mesh) is online with branded defaults. See the First-Run Bootstrap Overview for a detailed breakdown of each step.

Key Directories

  • CATACLYSM_STUDIOS_INC/ – Provisioning bundles and infrastructure automations for homelab and field hardware, including unattended OS installs, Jetson bootstrap scripts, and ready-to-run Docker stacks that mirror the production mesh topology.
  • docs/ – High-level strategy, architecture, and integration guides for the overall PMOVES ecosystem. See also pmoves/docs/ENVIRONMENT_POLICY.md for the single-file environment policy and Jellyfin host-mount instructions.
  • pmoves/ – The primary application stack with docker-compose definitions, service code, datasets, Supabase schema, and in-depth runbooks for daily operations and advanced workflows.
  • pmoves/contracts/solidity/ – Hardhat workspace prototyping Food-USD / GroToken governance flows with automated tests that model staking, quadratic voting, and group-buy execution.
  • pmoves/ui/ – Next.js + Supabase Platform Kit workspace for the upcoming web UI; reuses pmoves/.env.local so frontend hooks can target the same Supabase CLI stack.

Essential Documentation

  • Claude Code CLI Integration – TAC integration with custom slash commands, security hooks, and PMOVES-aware context for AI-assisted development.
  • Testing Strategy – Comprehensive testing guide covering smoke tests, functional tests, and end-to-end validation workflows.
  • PMOVES Stack README – Quickstart environment setup, service inventory, and Codex bootstrap steps for running the orchestration mesh locally.
  • Local Tooling Reference – One-stop index for environment scripts, Make targets, Supabase workflows, smoke tests, and provisioning helpers.
  • Supabase Service Guide – CLI vs compose expectations, realtime wiring (supabase start --network-id pmoves-net), and how PMOVES consumes PostgREST/Realtime in both local and self-hosted deployments.
  • PMOVES Docs Index – Curated entry points into the pmoves-specific runbooks covering Creator Pipeline, ComfyUI flows, reranker configurations, and smoke tests.
  • UI workspace bring-up – Next.js + Supabase quickstart (npm/yarn commands, env loading from pmoves/.env.local, Supabase CLI prerequisites).
  • Service Docs Index – Per-service guides (overview, compose/ports, runbooks, smoke tests, and roadmap alignment).
  • External Integrations Bring-Up – Wger, Firefly III, Open Notebook, and Jellyfin commands, token wiring, and port overrides for make up-external.
  • Architecture Primer – Deep dive into mesh topology, service responsibilities, and evolution of the orchestration layers.
  • Complete Architecture Map – Full-fidelity view of the latest integration mesh, including data planes and edge deployments.
  • Multi-Agent Integration Guidelines – Operational patterns for coordinating Agent Zero, Archon, and automation hubs across environments.
  • Archon Updates for PMOVES – What changed in the October 2025 Archon bundle, how to wire the Supabase CLI stack, and the MCP/NATS expectations.
  • Make Targets Reference – Command catalog for starting, stopping, and tailoring compose profiles (core data plane, media analyzers, Supabase modes, and agent bundles).
  • Single-User (Owner) Mode – Personal-first operation without login prompts; boot-JWT auto-auth, owner chip in the UI, and security notes.
  • Production Readiness Report – Feb 2026 audit of service health, security posture, and deployment readiness.
  • Codex + Copilot Review Workflow – How to combine the Codex CLI reviewer with GitHub Copilot's PR assistant, including token setup and evidence logging expectations.

CHIT & Geometry Documentation

Compressed Hierarchical Information Transfer (CHIT) and the Geometry Bus are core to how PMOVES.AI encodes, routes, and decodes structured knowledge across the agent mesh.

See pmoves/docs/PMOVESCHIT/ for the full 19-document collection including decoder specifications, neural-network notebooks, audit reports, and UI design specs.

Rooms on a Stage

PMOVES' Metal-Organic Framework architecture manifests operationally as rooms on a stage — a living topology (not a frozen specification) with three layers:

  • Rooms are the agent-owned entrypoints: infra fabric, field control, voice studio, workstation. Each room binds services, skills, and notebook state.
  • Stage is the lifecycle state per room: rehearsallivereviewarchive. Rooms transition through these states as work progresses.
  • Suits are the runtime/persona bindings layered onto rooms — upstream Agent Zero as the external baseline, PMOVES hardened overlays as the custom fit, voice/theme/persona as the visible styling layer.

P7 (Pinokio 7) is the evolving room-aware stage manager: it reads the room catalog, selects the appropriate room profile for a given workload, loads the correct suit, and manages stage transitions. P7's NATS subjects (p7.nats.launch, p7.nats.session) are the control plane for room entry and lifecycle. As the platform grows, P7's tree adds new rooms, stage states, and suit types — the topology is designed to expand.

Rooms own presentation and session ergonomics; the notebook plane owns durable memory. The Room Manifest Contract defines the interface; pmoves/config/rooms/catalog.json is the canonical seed catalog. See Rooms on a Stage Overview for the end-to-end model and AGNOTE4482 for the full P7 specification.

Room Index

🏗️ Z890 Infra Fabric Room

Purpose: Infrastructure room for topology, service health, secrets discipline, and operator bring-up. Profile: z890-infra · Agent: z890-claude · Stage: live · Manifest: z890-infra.room.fabric.json

  • pmoves/services/node-registry/ — Multi-host node discovery and health reporting
  • pmoves/services/resource-detector/ — Hardware capability detection (GPU, CPU, memory)
  • pmoves/services/work-marshaling/ — Distributed task scheduling and work queue management
  • pmoves/services/gpu-orchestrator/ — GPU resource allocation and scheduling
  • pmoves/services/vllm-orchestrator/ — vLLM inference server lifecycle management
  • pmoves/services/model-registry/ — Central model catalog with version tracking
  • pmoves/services/tensorzero-config-api/ — Dynamic TensorZero model configuration API
  • pmoves/services/benchmark-runner/ — Automated performance benchmarking harness
  • pmoves/services/nats-echo/ — NATS message debugging and replay tool
  • pmoves/services/analysis-echo/ — Analysis pipeline event echo and auditing
  • pmoves/services/evoswarm/ — Evolutionary swarm optimization coordinator

🔭 4090 Field Control Room

Purpose: Scout/control room for review, topology, handoff, and notebook-backed triage. Profile: 4090-field · Agent: 4090-claude · Stage: live · Manifest: 4090-field.room.control.json

  • pmoves/services/agent-zero/ — MCP bridge + decision engine (ingests Supabase + CHIT events)
  • pmoves/services/archon/ — Agent builder/knowledge management with Supabase CLI realtime + NATS clients
  • pmoves/services/deepresearch/ — Tongyi DeepResearch bridge with OpenRouter/local modes plus Open Notebook mirroring
  • pmoves/services/supaserch/ — Multimodal holographic deep research orchestrator
  • pmoves/services/graph-linker/ — Knowledge graph linking and entity relationship management
  • pmoves/services/session-context-worker/ — Session context aggregation for multi-turn agent conversations
  • pmoves/services/gateway-agent/ — Unified API gateway with agent-aware routing
  • pmoves/services/botz-gateway/ — Skills marketplace gateway for BoTZ agent capabilities
  • pmoves/services/agentgym-rl-coordinator/ — Reinforcement learning coordinator for agent skill training
  • pmoves/services/consciousness-service/ — Agent self-model and meta-cognitive state tracking
  • pmoves/services/chat-relay/ — Agent-to-agent and agent-to-user chat relay
  • pmoves/services/messaging-gateway/ — Multi-channel messaging gateway
  • pmoves/services/a2ui-nats-bridge/ — Agent Zero UI to NATS event bridge
  • pmoves/services/open-notebook/ — Streamlit UI + SurrealDB API (container ports 8502/5055 per upstream; host defaults map to :8503 UI and :5055 API, override with OPEN_NOTEBOOK_*_PORT) for research assets and MCP notebooks. Doc: pmoves/docs/services/open-notebook/
  • pmoves/services/notebook-sync/ — Bridges Open Notebook datasets into Supabase and LangExtract flows
  • pmoves/services/retrieval-eval/ — Retrieval benchmarking, relies on Supabase + hi-rag

🎙️ 5090 Voice Studio

Purpose: Voice-first room for TTS, media pipelines, notebook-backed iteration, and audition workflows. Profile: 5090-voice · Agent: 5090-claude · Stage: live · Manifest: 5090-voice.room.studio.json

  • pmoves/services/flute-gateway/ — Multimodal voice communication layer (HTTP :8055, WebSocket :8056) with Pipecat integration and prosodic synthesis
  • pmoves/services/vibevoice-realtime/ — Real-time voice synthesis service
  • pmoves/services/pmoves-yt/ — YouTube ingest; publishes geometry packets after segmentation
  • pmoves/services/channel-monitor/ — External content watcher; triggers ingestion on new uploads
  • pmoves/services/publisher/ — Discord & Jellyfin publisher with geometry-aware payloads
  • pmoves/services/publisher-discord/ — Dedicated Discord notification bot for ingest/summary events
  • pmoves/services/jellyfin-bridge/ + pmoves/docs/services/jellyfin-ai/ — Media sync bridging Jellyfin metadata into Supabase + Discord publisher
  • pmoves/services/{presign,render-webhook,extract-worker,langextract,media-audio,media-video,pdf-ingest,comfy-watcher,comfyui} — Supporting ingestion, extraction, and media tooling

💻 5090 KiloCode GLM Workstation

Purpose: GPU inference specialist node running KiloCode GLM on the GLM Coding Plan. Shares the 5090 with Claude Code and Codex. Profile: 5090-kilocode · Agent: 5090-kilocode · Stage: rehearsal · Manifest: 5090-kilocode.room.studio.json

  • pmoves/services/tensorzero-config-api/ — Dynamic TensorZero model configuration API (shared with Infra room)
  • pmoves/services/gpu-orchestrator/ — GPU resource allocation and scheduling (shared with Infra room)
  • pmoves/services/model-registry/ — Central model catalog with version tracking (shared with Infra room)
  • pmoves/services/retrieval-eval/ — Retrieval benchmarking (shared with Field Control room)

Cross-Cutting: Geometry & CHIT Fabric

These services form the knowledge and geometry substrate available to all rooms via the NATS event bus:

  • pmoves/services/hi-rag-gateway-v2/ — v2 gateway (CPU :8086, GPU :8087). Handles /geometry/*, jump, decode, calibration, Supabase realtime warmups, and CGP persistence.
  • pmoves/services/hi-rag-gateway/ — v1 legacy gateway (host :8089). Minimal CHIT endpoints for backward compatibility.
  • pmoves/services/gateway/ — Experimental CHIT UI/API for live geometry visualisation and WebRTC broadcast.
  • pmoves/services/mesh-agent/ — Geometry mesh bridge; signs and republishes geometry.cgp.v1 across deployments.
  • pmoves/services/evo-controller/ — Geometry tuning controller; reads CGPs from Supabase, emits tuning capsules back into the bus.
  • pmoves/services/tokenism-simulator/ — Token geometry simulation and visualization.

Cross-Cutting: Orchestration & External Data

  • pmoves/services/n8n/ — Workflow orchestrator; health/finance webhooks emit CGPs via hi-rag v2.
  • pmoves/services/wger/ — Health metrics ingest (paired with Supabase tables + health.weekly.summary.v1 CGPs).
  • pmoves/services/firefly-iii/ — Personal finance ingest; finance flows create finance.monthly.summary.v1 CGPs.

See each directory's README for ports, Make targets, and geometry notes. New integrations reference external repositories under integrations-workspace/ and the setup steps captured in pmoves/docs/EXTERNAL_INTEGRATIONS_BRINGUP.md.

Dashboards & UIs (local defaults)

By Room

Z890 Infra Fabric Room:

4090 Field Control Room:

5090 Voice Studio:

5090 KiloCode GLM Workstation:

Default access and operator credentials

  • Supabase operator is provisioned by make supabase-boot-user (also run by make first-run). The command writes values to pmoves/env.shared and pmoves/.env.local:

    • SUPABASE_BOOT_USER_EMAIL, SUPABASE_BOOT_USER_PASSWORD, SUPABASE_BOOT_USER_JWT.
    • The PMOVES UI auto-authenticates with NEXT_PUBLIC_SUPABASE_BOOT_USER_JWT so most routes won't prompt for a password. If you need to log in manually, use the email/password above from your env files.
  • Jellyfin uses the LinuxServer image defaults. After first boot, confirm the admin user and API key in pmoves/env.jellyfin-ai or via the Jellyfin UI (Settings → Dashboard). Update JELLYFIN_API_KEY and JELLYFIN_USER_ID in pmoves/env.shared if you rotate.

  • Wger and Firefly are brought up with PMOVES-branded defaults sourced from pmoves/env.shared (see pmoves/docs/FIRST_RUN.md "Seeded & Branded Defaults" for the exact initial usernames and emails).

  • Open Notebook's UI password also serves as its API bearer token; keep OPEN_NOTEBOOK_API_TOKEN identical to OPEN_NOTEBOOK_PASSWORD so CLI helpers and agents work against the same branded login (see pmoves/docs/services/open-notebook/README.md).

  • For a full list of seeded branded logins and where they come from, see:

    • pmoves/docs/FIRST_RUN.md (Seeded & Branded Defaults)
    • docs/SECRETS.md (Secret Management Playbook)
  • Creator bundle: see pmoves/creator/ for installers, tutorials, and ComfyUI workflows supporting WAN Animate, Qwen Image Edit+, and VibeVoice TTS. Key guides include:

  • Creator Pipeline Runbook – Current status of n8n automations (health/finance live, creative flows staging) plus geometry mapping and persona playback prep.

Developer Tools & Testing

Claude Code CLI Integration (TAC)

PMOVES.AI integrates deeply with Claude Code CLI through the .claude/ directory, providing context-aware development assistance:

Key Features:

  • Always-on context (.claude/CLAUDE.md) - Makes Claude aware of PMOVES architecture, services, and patterns
  • Custom slash commands - Quick access to production services:
    • /health:check-all - Verify all services are running
    • /search:hirag "query" - Query Hi-RAG v2 knowledge base
    • /agents:status - Check Agent Zero orchestration status
    • /deploy:smoke-test - Run integration test suite
  • Security hooks - Pre-tool validation blocks dangerous operations (rm -rf /, DROP DATABASE, etc.)
  • Observability hooks - Post-tool events published to NATS (claude.code.tool.executed.v1)

Quick Start:

# Use Claude Code CLI with PMOVES context
claude

# Try a custom command
/health:check-all

# Query knowledge base
/search:hirag "How does Agent Zero work?"

Learn more: .claude/README.md | Developer Context

Testing Infrastructure

PMOVES.AI employs comprehensive testing from smoke tests to end-to-end workflows:

1. Smoke Tests (30-60 seconds) - Quick health validation:

cd pmoves
make verify-all          # Full smoke test suite
make smoke               # Core services only
make smoke-gpu           # GPU-enabled services
make archon-ui-smoke    # Archon agent services
make deepresearch-smoke  # Research orchestration

2. Functional Tests (2-5 minutes) - Feature validation:

# Run all functional tests
cd pmoves/tests
./run-functional-tests.sh

# Individual test categories
./run-functional-tests.sh --category retrieval
./run-functional-tests.sh --category agents
./run-functional-tests.sh --category ingestion

3. Integration Tests - End-to-end workflows:

# NATS event flow validation
make test-nats-flow

# Media pipeline validation
make test-media-pipeline

# Agent coordination validation
make test-agent-mesh

Coverage Map:

  • Service health endpoints (Qdrant, Neo4j, Meilisearch, Supabase)
  • Retrieval accuracy (Hi-RAG v2 reranking, embedding generation)
  • Agent coordination (Agent Zero MCP, Archon prompts, NATS routing)
  • Media processing (YouTube ingestion, Whisper transcription, YOLO analysis)
  • Observability (Prometheus metrics, Grafana dashboards, Loki logs)

Learn more: docs/testing/TESTING.md

Development Workflow

Typical developer flow:

  1. Environment setup - make first-run or make bootstrap
  2. Service verification - make verify-all (smoke tests)
  3. Development - Use Claude Code CLI with custom commands
  4. Testing - ./run-functional-tests.sh before commits
  5. Monitoring - Grafana dashboard at http://localhost:3000

Best practices:

  • Use /health:check-all before starting work to verify service status
  • Leverage existing services (Hi-RAG, SupaSerch, Agent Zero) - don't duplicate
  • Publish events to NATS for async coordination
  • Follow observability patterns (expose /healthz and /metrics)
  • Run smoke tests after infrastructure changes
  • Use security hooks to prevent dangerous operations

Build Status & Security

Security posture: 29 open CodeQL alerts (2 critical, 22 high, 5 medium) — triaged into 7 remediation groups. PRs #651/#653/#654 resolved 36 alerts, but the Hardened branch scope re-surfaced 29 on a broader CodeQL scan. The 2 critical alerts are SSRF findings in Hi-RAG gateway services requiring URL allowlisting.

Pre-production blockers (6 remaining): AB-1 (A2UI nested gitlink), AB-3 (GHCR triggers), AB-4 (real credentials), AB-5/AB-6 (runtime validation, deferred until AB-4), AB-7 (PBKDF2 iteration bump). See the Production Audit Dashboard for full details and resolution sequence.

CI gates (all enforced on PRs to main):

  • CodeQL Advanced — Static analysis for JS/Python/Go vulnerabilities
  • CHIT Contract Check — Schema validation for geometry packet contracts
  • SQL Policy Lint — Migration and policy validation
  • Docker Hardening Validation — Container security baseline enforcement
  • Integration Contract — Cross-service API contract verification
  • Python Tests — Unit and integration test suite

Key security PRs:

  • #651 – Initial CodeQL alert triage (36 → 12)
  • #653 – Remaining CodeQL fixes (12 → 6)
  • #654 – Final 6 alerts resolved across gateway and YT services
  • Note: 29 alerts re-surfaced on the Hardened branch due to expanded CodeQL scan scope

Docker build reliability: All core services build successfully. See Build Fixes Documentation for historical context on DeepResearch, FFmpeg-Whisper, and environment file fixes.

Production readiness: See the Production Audit Dashboard for the consolidated audit covering active blockers, CodeQL triage, static audit layers, and resolution sequence. Historical context: Feb 2026 Readiness Report.

Getting Started

  1. Bootstrap the stack – For brand-new machines run make first-run. For incremental setup follow the environment and container launch instructions in the pmoves/README.md: place overrides in pmoves/.env.local, run make bootstrap to capture credentials, make up to start the core services, and make bootstrap-data to apply Supabase SQL, seed Neo4j, and load the demo Qdrant/Meili corpus before smoke testing.
  2. Review orchestration flows – Use the Make Targets Reference for day-to-day compose control, and consult the architecture and multi-agent guides in /docs for how Agent Zero, Archon, and supporting services communicate across the mesh.

Initial Setup & Tooling Flow (manual path)

  1. Environment bootstrap – Walk through pmoves/README.md to provision runtime prerequisites, seed pmoves/env.shared, and populate secrets. Use make bootstrap (wrapping python -m pmoves.scripts.bootstrap_env) when you need finer control, or invoke python3 -m pmoves.tools.mini_cli bootstrap --accept-defaults to script the same flow alongside the provisioning bundle. Both paths update env.shared, .env.generated, .env.local, and the auxiliary env.*.additions files consumed by Compose and the UI launcher.
  2. Supabase realtime alignment – Follow the Supabase Service Guide to start the CLI stack with supabase start --network-id pmoves-net (run this before accepting Supabase prompts in make bootstrap) and mirror the websocket endpoint (SUPABASE_REALTIME_URL=ws://host.docker.internal:65421/realtime/v1). This matches our self-hosted Supabase deployments.
  3. UI workspace bring-upcd pmoves/ui then npm install (or yarn install). The Next.js app loads Supabase creds from pmoves/.env.local and expects the Supabase CLI stack (make supa-start + make supa-status) before running npm run dev.
  4. Tooling cheatsheet – Keep Local Tooling Reference handy for Make targets, smoke tests, and environment scripts (env_setup, flight-check, smoke).
  5. Provisioning & hardware targets – Browse CATACLYSM_STUDIOS_INC/ for automated OS images, Jetson bootstrap bundles, and pmoves-net Docker stacks ready for edge hardware.

Contributors

POWERFULMOVES

3,095 commits

hunnibear

677 commits

dependabot[bot]

138 commits

Barathicite

15 commits

Languages

Python

46.1%

HTML

18.8%

Jupyter Notebook

14.2%

TypeScript

5.4%

Shell

4.8%

C#

2.7%

JavaScript

1.8%

Makefile

1.6%

PLpgSQL

1.6%