Every system, no matter how self-contained it aspires to be, stands on the shoulders of other systems. The Consilience is no exception. What follows is not a bibliography — it is the working index. Every tool, every service, every repository that the platform depends on, reaches into, or opens outward to. A student who reads this chapter should be able to find everything that matters, in the order it matters.
*
The Consilience's own surfaces
The open-source monorepo lives at github.com/kaOS-Intelligence/kaOS. Every package, every app, every configuration file is public. The branch is master. The license is MIT for the reusable packages and a custom protective license for the sovereign runtime — usage is free, redistribution requires attribution, and the Sacred Boundary files carry additional protection.
@kaos/claw is published to npm at npmjs.com/package/@kaos/claw. Installing it with npx @kaos/claw launches a working MCP server connected to the Consilience library API. It is the canonical starting point for anyone who wants to build tools against the platform.
The kaOS Intelligence website at kaosintelligence.com carries the public face of the organization — what it builds, what it believes, how to reach it. The Chronicle — the serialized literary epic about the Aeonari civilization — lives at its own surface, reached through the library app. The Consilience Library is the growing collection of architectural documents, design specifications, and writing guides that describe how the system was built and why it was built that way. The PostHog operator guide, at docs/posthog-operator-guide.md in the repository, explains the analytics taxonomy, the event schema, and the preset query system in language meant for operators, not engineers.
*
External tools — the model layer
The model ecosystem that powers the platform is deliberately diverse. No single provider. No single architecture. The principle is sovereignty: every dependency should be replaceable without rewriting the system.
Hugging Face (huggingface.co) is the hub. Models, datasets, training jobs, evaluation suites, and the community that builds them. The platform uses Hugging Face for model discovery, for the BGE-M3 embedding model that powers every semantic search, for the voice models that feed the TTS pipeline, and for the training infrastructure that runs fine-tuning jobs. The kaos-huggingface skill documents every integration path.
DeepSeek (deepseek.com) provides the primary cloud fallback model — DeepSeek V4 Pro — and the sub-agent workhorse DeepSeek V4 Flash. The API is Anthropic-compatible, which means the same client code that talks to Anthropic can talk to DeepSeek with a URL change and a different API key.
Anthropic (anthropic.com) provides Claude Opus as the secondary cloud fallback. It is not the default because the economics favored DeepSeek for normal usage, but it remains in the chain because it has been reliable, and redundancy is protection.
Cohere (cohere.com) provides the embed-v4.0 model for semantic search, the Rerank endpoint for search refinement, and the Command R models as an alternative text generation path. The central AI package abstracts all three into a single provider interface.
Gemini (ai.google.dev) provides the Laomedeia voice model for cloud TTS fallback and Gemini Flash for specific sub-agent tasks.
*
External tools — the infrastructure layer
Ollama (ollama.com) runs the local models on the Mac Studio. It loads the Conductor's dense 27.8B Q8 body, the BGE-M3 embedder, the vision-grounding specialist, and the code-specialist jacket. It is the substrate — the thing that actually runs the weights. Without it, the platform would be entirely cloud-dependent.
LiteLLM (litellm.ai) proxies every model request through a single local gateway at port 4000. It logs cost, latency, tokens, and model selection. It routes between local Ollama models and cloud providers transparently. It is the platform's accounting layer for intelligence — the place where every turn is measured and every dollar is tracked.
Tailscale (tailscale.com) provides the encrypted tunnel between the VM in Montreal and the Studio in Vancouver. The Tailscale Funnel makes the LiteLLM proxy reachable from the public internet without opening a single port. It is the network layer's answer to the sovereignty question: a private machine, on a private network, reachable only by those who know the address, secured by WireGuard.
Google Cloud Platform runs the lightweight VM that hosts the OpenClaw gateway, the Telegram bot bridge, and the email agent. The instance is e2-small — two virtual CPUs, two gigabytes of RAM. It does almost nothing except route packets, and that is by design. The heavy compute lives on the Studio.
ToolPiper (toolpiper.app) runs as a macOS application on the Studio, providing voice synthesis through its MCP server. The broker connects to it via the same protocol everything else uses — typed tools, structured requests, audio data in response.
Whisper.cpp (github.com/ggerganov/whisper.cpp) handles speech-to-text locally on the Studio. Voice notes, phone calls, and dictation all route through it. It runs on the same machine as the Conductor's body, sharing the same GPU.
*
External tools — the platform layer
Vercel (vercel.com) deploys the web surfaces — the library app, the API, the admin dashboard, the publisher gateway, the Bible app, the Telegram mini-app, and the marketing site. Seven projects, one monorepo, each with its own root directory and environment variables.
Supabase (supabase.com) provides the PostgreSQL database with pgvector for semantic search, authentication for the non-Clerk surfaces, and file storage. It is the platform's persistence layer. The database is production. The rule is read-only for agents.
Clerk (clerk.com) handles user authentication across all web surfaces with multi-tenant support, organization management, and webhook-driven identity sync.
Stripe (stripe.com) processes payments — subscriptions, one-time purchases, credit top-ups, invoicing. The integration lives in the shared payments package and is consumed identically by every surface that handles money.
Resend (resend.com) delivers transactional email — the daily digests, the reading reminders, the account notifications — using React Email templates rendered server-side and sent through Resend's API.
PostHog (posthog.com) captures product analytics, feature flags, experiments, session recordings, error tracking, and LLM observability. The platform's self-knowledge lives here.
Sentry (sentry.io) captures errors with full stack traces, user context, and release information. It is the platform's early warning system.
Slack (slack.com) carries team notifications — deployment status, error spikes, agent pipeline summaries.
Figma (figma.com) holds the design system components, tokens, and screens. The platform reads from it and writes to it through the MCP layer.
Browserbase (browserbase.com) provides headless browser infrastructure for tasks that need to see a rendered page — visual regression testing, content extraction, accessibility auditing.
GitHub (github.com) hosts the repository, runs CI/CD through Actions, and manages issues and pull requests.
*
Everything listed here is reachable. Everything is documented. Nothing requires a special relationship or a private API key that only the Architect holds. The platform is open in the practical sense: the tools it uses are the tools anyone can use, the protocols it speaks are the protocols anyone can implement, and the code it runs is the code anyone can read. The only gate is the work of understanding, and that work is what this student pack exists to reduce.