Governed adapters. Every capability evidence-wrapped.

Governed HTTP client with URL allowlist

Wrap Model Context Protocol (MCP) server tools as governed CHP capabilities

Governed file read/write/list with path allowlist

Git repository inspection and operations

Report and update the CHP host runtime on a node (governed self-update)

Run any capability as a polled background job

Runtime credential injection from env/file/memory backends

Static and runtime violation checker for capability implementations

Tailscale mesh networking for multi-host CHP clusters

Queryable audit log over the host evidence store

Codebase capability scanner — surfaces candidate adapters & capabilities (powers onboarding)

Governed capability discovery registry over the host catalog

Run the offline steward fleet (emission/conformance/doc/onboarding/demand) on-demand

GraphQL queries and mutations as governed capabilities

gRPC service discovery and unary call invocation

Enterprise Kafka event stream operations

NATS messaging (pub/sub/request-reply) as governed capabilities

Any REST API described by an OpenAPI spec as governed CHP capabilities

Inbound webhook signature verification and event normalization

Document search, retrieval, and management

File extraction and format conversion

LLM and capability evaluation rubrics

Local HuggingFace artifact consumption — pull, pipeline, embed, tokenize, dataset, cache

Bridge CHP capabilities to LLM tool-calling APIs (OpenAI, Anthropic, generic)

Local LLM inference via Ollama (primary) or llama.cpp (fallback)

Governed scoped key-value memory for agents

Apple Silicon native text generation via a local mlx_lm server

FastContext repo scout subagent

SGLang inference server (CUDA/Linux primary, CPU fallback)

Code-writing meta-agent (smolagents) whose tools are CHP capabilities

Metal-accelerated text embeddings via a local Text Embeddings Inference server

Semantic embedding search with injectable backend

Apple Silicon native text generation via a local vLLM server

Run package test suites as governed evidence-producing invocations

GitHub repository, pull request, issue, and CI inspection as governed capabilities

Radicle p2p code forge (sync, patch, issue, repo identity)

CHP adapter — WireMock admin API as governed capabilities (openapi-backed; test the factory)

Docker/OCI container execution

Kubernetes cluster management as governed capabilities

Govern macOS launchd (LaunchAgent) services

Governed subprocess/CLI execution

Release pipeline operations (bump, sync, tag, publish)

Governed finite state machines

Durable workflow execution tracking with injectable backend

AsyncAPI spec parsing and channel-based messaging

Compose capabilities into named reusable workflows

Governed work handoff and delegation lifecycle

Organizational hierarchy and people lookup

Observable agent cognition (planning and reflection)

Browser/UI automation (RPA)

Risk assessment and guardrail evaluation

CHP adapter — anchore.io API as governed capabilities (openapi-backed, generated)

AWS infrastructure operations as governed capabilities

Azure infrastructure operations as governed capabilities

Google Cloud Platform infrastructure operations

Raspberry Pi physical-world access (GPIO, camera, system stats)

Synology DSM file operations, tasks, and Container Manager

Vercel deployment platform as governed capabilities

Outbound email via SMTP or Resend

Matrix messaging + human-in-the-loop approvals as governed CHP capabilities (composes through chp.adapters.http)

Conversation transcript storage across JSONL and remote CHP backends

Multi-channel notification routing

Operate a self-hosted Postiz social scheduler via its Public API

Slack messaging, interactive approvals, and signed-interaction ingest

Calendar event management

Structured form definition, submission, and retrieval

Jira Cloud issue management

Knowledge graph with injectable graph backend

Linear project management via GraphQL

Notion pages and databases as governed capabilities

Governed PostgreSQL query, mutation, and schema introspection

Per-agent/per-user preference storage

Sovereign LLM-driven structured web/document extraction (self-hosted ScrapeGraphAI)

CSV and spreadsheet read/write

Multi-hop transformation path planner

effect pre-authorization (single-use, action-bound approval grants)

Machine-to-machine OAuth2 client credentials and JWT validation

governed headless-browser navigation with a network-enforced domain allowlist

sovereign frontier (MoE) text generation via a local colibri server (github.com/JustVugg/colibri)

bounded recursive self-improvement (propose + gate new capabilities)

on-mesh model eval with signed, portable scorecards

the recursive flywheel's decision steps (curate + promotion gate + evaluate_and_gate)

sovereign code/artifact knowledge graph via graphify (BYO mesh backend)

hardware-aware LLM model fit, ranking, and discovery (wraps the llmfit CLI)

personal notes over a markdown/Obsidian vault

rootless isolation-boundary command execution (evidence-recorded)

host and lifecycle-manage long-running services on the mesh

emit agent/LLM runs as standard OpenTelemetry GenAI spans

bounded sovereign web search via a self-hosted SearxNG