THE ORCHESTRATOR

Kompile CLI

One command-line harness to bootstrap, configure, and run your entire AI stack. Create projects that crawl your data and compile knowledge. Over 100 native commands that orchestrate models, data pipelines, servers, and agents from your terminal — against any provider.

The Single Entry Point

The CLI is how you interact with every part of the Kompile stack. It bootstraps runtimes, manages model lifecycles, launches servers, runs chat sessions, and orchestrates data pipelines — all through a single kompile command.

100+

Native Commands

1 Binary

Single Install

All Providers

One Unified Interface

Kompile Projects

A Kompile project is a self-contained unit that crawls your data and compiles it into knowledge your AI can act on. Each project bundles its own configuration, pipelines, knowledge graphs, and model assignments.

Crawl & Compile

Point a project at your data sources — Confluence, Jira, Slack, local files, web pages — and it automatically crawls, chunks, embeds, and indexes everything into structured knowledge.

Scoped Configuration

Each project carries its own model assignments, vector store indexes, pipeline definitions, and knowledge graph. Run multiple projects with different providers and configurations on the same machine.

Knowledge Graphs

Every project builds a knowledge graph from its crawled data — entities, relationships, communities, and fact sheets. Query the graph directly, run algorithms, or use it for GraphRAG.

kompile init scaffolds a new project → kompile crawl crawls your sources → kompile chat starts answering questions from compiled knowledge

Chat

Chat natively or bring your own agent. Kompile gives you multiple ways to interact with your compiled knowledge — each with the full power of the stack behind it.

Native Chat

Run kompile chat for an interactive session against any configured LLM — local models, OpenAI, Anthropic, Gemini, or custom endpoints. Full access to every Kompile tool: RAG search, knowledge graphs, model management, and data crawlers.

Passthrough Mode — Your Agent, Our Tools

Run kompile passthrough claude and Kompile launches your preferred CLI agent with every Kompile tool injected as MCP tools. Works with claude, codex, gemini, qwen, opencode. Zero workflow disruption.

Cross-Agent Resume

Every conversation is stored in a unified transcript store. Resume any previous session where it left off — even with a different agent. Start in Claude, pick up in Gemini, continue in Kompile chat. The context follows you, not the provider.

Transcript Search & Import

Full-text search across every past conversation — find sessions by keyword, date, agent, or topic. Import conversation history from Claude, Codex, Gemini, or Qwen into one unified store. Never lose context again.

MCP Server Mode

Run kompile mcp-stdio and the CLI becomes a Model Context Protocol server. Claude Desktop, Cursor, VS Code, and any MCP client can use your full Kompile stack as a tool provider.

Command Reference

Commands grouped by what they do. Every operation in the stack is accessible from the terminal.

Build & Install

kompile install

Bootstrap the full Kompile stack — downloads runtimes, configures paths, and verifies dependencies.

kompile build

Compile the entire platform or individual modules from source with automatic dependency resolution.

kompile update

Pull the latest version of all Kompile components and rebuild in place.

kompile init

Initialize a new Kompile project with config scaffolding, directory structure, and default pipelines.

Model Management

kompile model download

Fetch models from HuggingFace or remote registries with resumable downloads and progress tracking.

kompile model convert

Convert between formats — ONNX, TensorFlow, Keras, GGUF/GGML, SameDiff — with automatic graph optimization.

kompile model list

List all models in the local registry with status, format, size, and device assignment.

kompile model promote / replace / delete

Full lifecycle management — stage, promote to production, hot-swap, or remove models.

Archive & Air-Gap

kompile archive export

Package models into .karch archives with manifests and SHA-256 checksums for air-gapped transfer.

kompile archive import

Load .karch archives into the local registry — works fully offline behind any firewall.

kompile archive publish

Push .karch archives to a remote registry for team distribution.

kompile archive list

Browse available archives in local and remote registries.

Chat & Interaction

kompile chat

Interactive chat session against any configured LLM — local models, OpenAI, Anthropic, or custom endpoints — with full access to all Kompile tools.

kompile passthrough

Keep using your existing CLI (Claude, Codex, Gemini, Qwen, OpenCode) but with all Kompile tools injected as MCP tools. Your agent, our tools — zero workflow disruption.

kompile mcp-stdio

Launch as an MCP (Model Context Protocol) server over stdio for integration with Claude, Cursor, and other MCP clients.

kompile api-agent

Run a preconfigured API agent against any OpenAI-compatible endpoint (Ollama, vLLM, LM Studio).

kompile cli-agent

Orchestrate external CLI agents (Claude CLI, Codex CLI, Gemini CLI, Qwen CLI) through a unified interface.

kompile transcript search

Full-text search across all past conversation transcripts — find previous sessions by keyword, date, agent, or topic.

kompile transcript import

Import conversation transcripts from external agents (Claude, Codex, Gemini, Qwen) into the unified transcript store.

kompile resume

Resume a previous conversation where it left off — across any agent. Start with Claude, pick up with Gemini, continue in Kompile chat.

Projects

kompile init

Initialize a new Kompile project with config scaffolding, directory structure, default pipelines, and model configuration.

kompile project

Run project-level operations — knowledge crawls, VLM OCR presets, and project-scoped configuration.

kompile knowledge

Manage the project knowledge base — import markdown, crawl sources, and build knowledge graphs scoped to the current project.

kompile graph

Full knowledge graph operations — add/delete nodes and edges, traverse, run algorithms, detect communities, query, import/export, and launch an interactive graph shell.

Data Pipelines

kompile ingest

Run document ingestion pipelines — load, chunk, embed, and index documents into your vector store.

kompile crawl

Crawl data sources (Confluence, Jira, Notion, Slack, web) and feed into the ingest pipeline.

kompile index

Build or rebuild vector indexes across all configured vector stores.

kompile schedule

Set up recurring crawl and ingest jobs with cron-style scheduling.

Server & Config

kompile start

Launch the Kompile App server, Model Staging server, or both as managed subprocesses.

kompile config

View and edit configuration — model endpoints, vector stores, embedding providers, GPU assignments.

kompile status

Health check across all running services, loaded models, and GPU utilization.

kompile subprocess

Manage background services — start, stop, restart, and tail logs for any Kompile subprocess.

MCP Server & SDK Mode

The CLI doubles as an MCP server and an SDK entry point. Integrate Kompile into any MCP-compatible client or embed it as a subprocess in your own toolchain.

MCP-STDIO Server

Launch kompile mcp-stdio and the CLI becomes a Model Context Protocol server. Claude Desktop, Cursor, VS Code, and any MCP client can use your full Kompile stack as a tool provider.

Subprocess SDK

Embed Kompile as a managed subprocess in your own applications. The CLI handles process lifecycle, health checks, log streaming, and graceful shutdown — your code just sends commands.

Ready to compile your stack?

Get early access and start orchestrating from the terminal.

Request Early Access