policy-orchestrator

Development control plane for a 26-repo ecosystem. Centralizes policy enforcement, vector DB management, encrypted GitHub Pages deployment, build provenance, and cross-repo coordination.

Quick start

uv sync
devctl              # dashboard + cheat sheet
devctl health       # full system check

devctl

Run bare for live dashboard:

$ devctl

  devctl — control plane

  26 repos  ai-agents:6  infrastructure:6  legal:4  quant-finance:5  ...
  2,050,352 vectors  across 12 collections
  11 page sections  8 deployed  3 pending

  Commands

  Repos      status, list, audit, discover, inventory
  Data       db-status, search, embed, audit-vectors
  Pages      deploy-pages, audit-pages, verify-pages
  Secrets    secrets, validate-secrets
  Facts      log-fact, query-facts, log-feedback, query-feedback
  Build      provenance, benchmark, health
  Tools      sync, dashboard, readme, policy, ingest-sessions, search-sessions

  Quick Start

  Check everything           devctl health
  What's dirty?              devctl status --dirty
  Search all data            devctl search "query"
  Deploy pages from cwd      devctl deploy-pages --auto --verify --push
  Run benchmarks             devctl benchmark
  Scale projection           devctl benchmark --project=100000
  Vector DB overview         devctl db-status
  Security audit pages       devctl audit-pages
  Live page verification     devctl verify-pages --quick
  What generated a file?     devctl provenance show path/file.html
  Stale outputs?             devctl provenance stale
  Log a verified fact        devctl log-fact --fact "X" --source-type email --confidence verified --domain legal

  Full reference: docs/DEVCTL.md

What it manages

26 repos across 7 categories

Category	Repos	Key
Legal	div_legal, caseledger, legal-tax-ops, words_quantum_legal, morpheme-page	Case analysis, document intelligence
AI/Agents	cortex, puffin, llm-router, vector-lab, joel-knowledge, open-multi-agent-fork	Agent frameworks, LLM routing
Quant/Finance	vpin, alpha_research, ts_embed, cyfopt	Quantitative research
Infrastructure	policy-orchestrator, contacts, d72, docvec, gpu-workers, energy_texas	Control plane, vectordb, GPU compute
Creative/Math	Escher, darkgallery	Visualizations
Web/Portfolio	jthorvaldur.github.io, bulldogs	Public sites
Product	caseledger	Legal case management

2M+ vectors across 12 collections

Two Qdrant instances, all hybrid (dense BGE + sparse SPLADE).

devctl db-status    # colored overview with chunking info + coverage
devctl search "X"   # federated search across all collections

Collection	Points	Owner
case_docs	1.7M	caseledger
legal_docs_v2	239K	div_legal
claude_code_sessions	68K	policy-orchestrator
whatsapp_chats	19K	contacts
openai_chats	5.6K	div_legal
contacts	3.4K	contacts
claude_chats_ai	2K	contacts
+ 5 more

146 encrypted GitHub Pages

AES-256-GCM client-side encryption, 3 password zones.

devctl deploy-pages --auto --verify --push   # deploy from current repo
devctl audit-pages                           # security policy check
devctl verify-pages                          # live HTTPS decrypt test

Build provenance + benchmarks

devctl provenance show reports/timeline.html   # what generated this file?
devctl provenance stale                        # outputs needing regeneration
devctl benchmark                               # time all operations
devctl benchmark --project=100000              # project costs at scale
devctl health                                  # full system status

Architecture

INTENT.md                              <- root authority
  ├── policies/hard/                   <- ERROR on violation
  │   ├── secrets.md                   <- never commit secrets
  │   ├── git-main.md                  <- no force push
  │   ├── legal-data.md               <- legal data boundaries
  │   ├── pages-encryption.yaml       <- AES-GCM encryption rules
  │   └── quantization.yaml           <- int8 for >10K point collections
  ├── policies/soft/                   <- WARN only
  ├── registries/
  │   ├── repos.yaml                  <- 26 repos
  │   ├── vector-collections.yaml     <- 12 collections with chunking/coverage
  │   ├── pages.yaml                  <- 11 page sections with generators
  │   ├── providers.yaml              <- 6 LLM providers with task routing
  │   └── secrets.schema.yaml         <- key profiles + validation patterns
  ├── lib/
  │   ├── llm_router.py               <- single-file LLM provider router
  │   ├── provenance.py               <- build lineage tracking
  │   └── profiler.py                 <- operation timing + throughput
  ├── scripts/                        <- 20+ enforcement and utility scripts
  ├── docs/
  │   ├── DEVCTL.md                   <- full command reference
  │   ├── devctl-commands.yaml        <- machine-readable command registry
  │   ├── DEPLOYMENT.md               <- provider + Vast.ai + pages guide
  │   └── KEY_ACQUISITION.md          <- API key priority spec
  ├── templates/                      <- standard files synced to managed repos
  └── adr/                            <- architectural decision records

Hub-and-spoke: code stays distributed, governance is centralized. See adr/0001-control-plane-architecture.md.

Global tools

gai                 # repo status across all ~/GitHub/ (pub/priv, clean/dirty)
gai commit          # AI commit message + push for each dirty repo
gai pages           # GitHub Pages deployment status
gai providers       # LLM provider availability
gai vast            # Vast.ai GPU instance status
gai secrets         # key validation
gai env             # keys.zsh propagation check

Governing document

INTENT.md is the root authority. Core directive:

Maximize alignment with repository intent. Not output volume.

Full command reference

See docs/DEVCTL.md for all 27 commands with options, examples, and cross-references.

Machine-readable: docs/devctl-commands.yaml.

Managed by policy-orchestrator. Category: infrastructure. 53 commits, last updated 10 minutes ago.