# DODIL — comprehensive AI cloud > Dodil is an end-to-end AI cloud. Six products on one platform: **Ignite** (serverless apps + managed model inference), **DataK³** (a multi-dimensional data plane — objects, HTAP tables, vector, and graph in one bucket), **Scriptum** (a typed workflow & agent engine), **Git** (source control with PRs and CI checks), **Registry** (a Docker-compatible container registry), and **AppID** (sign-in and identity for the apps you build). This is the documentation hub at `docs.dodil.io`; each product has its own docs (and its own `llms.txt`). Every service is a **strongly-typed gRPC API** (`dodil..v1.`) auto-exposed over **HTTP/JSON**. The surface is **discoverable without credentials** via gRPC server reflection — an agent can map every service, method, and type before it has a token, then authenticate only to call. Auth is a Dodil **IAM** bearer token (OIDC client-credentials). ## Start here (hub) - [Overview](https://docs.dodil.io/): the platform and its services - [Platform Design](https://docs.dodil.io/platform-design): RPC-first + HTTP, and agent-friendly discovery via reflection - [CLI](https://docs.dodil.io/cli): install the unified `dodil` CLI; per-product groups (`dodil ignite|data|scriptum|git|registry|appid`) - [MCP server](https://docs.dodil.io/mcp): connect any AI assistant to Dodil; tools mirror the CLI across every product - [Remote server](https://docs.dodil.io/mcp/remote): the hosted, multi-tenant endpoint at `https://mcp.dodil.io` (streamable HTTP; paste the bare origin — no `/mcp`, no trailing slash). Sign-in through Dodil Connect: protected-resource metadata points at `https://connect.dodil.io` (OAuth 2.1, PKCE S256, client ID metadata documents, public clients; no client ID or secret to enter). The user chooses which organizations to grant; with several, each tool call passes `org` (`auth_orgs` lists them). Partner-only address with payment tools: `https://mcp.dodil.io/partner`. Nothing to install. - Per-client guides: [Claude](https://docs.dodil.io/mcp/claude) (Customize → Connectors → Add), [Claude Code](https://docs.dodil.io/mcp/claude-code) (`/plugin marketplace add dodilio/dodil-mcp` then `/plugin install dodil@dodil`, or `claude mcp add --transport http dodil https://mcp.dodil.io`), [ChatGPT](https://docs.dodil.io/mcp/chatgpt) (custom connector / app), [other clients](https://docs.dodil.io/mcp/other-clients) (Cursor, VS Code, Gemini CLI, Windsurf, Kiro, Codex, Kimi Code, Goose: coming soon) - [Privacy & permissions](https://docs.dodil.io/mcp/privacy-and-permissions): what the connector accesses, tool safety labels, revoking access. MCP Registry name: `io.dodil/dodil` - [Local server](https://docs.dodil.io/mcp/local): `dodil mcp` — stdio or HTTP, shared auth/config with the CLI, optional `--enable-exec` code mode - [Troubleshooting](https://docs.dodil.io/mcp/troubleshooting) ## Authentication (IAM) - [IAM Overview](https://docs.dodil.io/iam): identity and access for the platform - [Service Accounts](https://docs.dodil.io/iam/service-accounts): create a `client_id` + `client_secret` - [Get an Access Token](https://docs.dodil.io/iam/access-tokens): OAuth 2.0 client-credentials against `id.dodil.io`, realm `dodil` — examples in curl, Python, Node.js, Go. The token is your bearer token for every API. - [Roles and Policies](https://docs.dodil.io/iam/roles_and_policies): what a token is allowed to do (scopes, DRNs) ## Discover the API (reflection — no auth) - gRPC endpoints: `rpc.dodil.io:443` (Ignite, Scriptum) and `rpc.data.dodil.io:443` (DataK³ control plane). - `grpcurl rpc.dodil.io:443 list` and `grpcurl rpc.data.dodil.io:443 list` enumerate every service; `grpcurl describe ` shows exact request/response types. No token required for discovery. - DataK³'s DATA plane additionally speaks wire-adapter protocols — Postgres (`pg.uk-lon-1.dodil.io:5432`, `sslmode=require`), S3 (`object.uk-lon-1.dodil.io`), Bolt/Cypher (`bolt+s://bolt.uk-lon-1.dodil.io:7687`), Qdrant (`qdrant.uk-lon-1.dodil.io`), Pinecone (`pinecone.uk-lon-1.dodil.io`), GraphQL (`gql.uk-lon-1.dodil.io/graphql`) — stock clients work as-is; see https://docs.dodil.io/k3/connect - HTTP/JSON transcoding is available on each service — see a product's Conventions page for its HTTP base (most under `api.dodil.io`). ## Examples (end to end — CLI and code) - [Examples](https://docs.dodil.io/examples): cross-product walkthroughs, each built two ways (CLI for humans, API/SDK for apps & agents) - [RAG knowledge base](https://docs.dodil.io/examples/rag-knowledge-base): ingest → embed → vector search → LLM answer (DataK³ + Ignite Models) - [Deploy a microservice](https://docs.dodil.io/examples/deploy-a-microservice): ship a handler as an auto-scaling Ignite app and invoke it - [Agentic workflow](https://docs.dodil.io/examples/agentic-workflow): a Scriptum workflow that calls tools and branches on LLM decisions - [Semantic search](https://docs.dodil.io/examples/semantic-search): vector search over a DataK³ bucket ## Products - [Ignite](https://docs.dodil.io/ignite) — serverless apps (Compute), managed inference (Models, OpenAI/Cohere-compatible), remote image builds (Builds), and secrets. Full index: https://docs.dodil.io/ignite/llms.txt - [DataK³](https://docs.dodil.io/k3) — the multi-dimensional data plane: objects, HTAP tables (Delta Lake), vector, and graph in one bucket, queryable over S3/Postgres/Bolt/Pinecone/Qdrant wire adapters, automated by Scriptum-template pipelines. Full index: https://docs.dodil.io/k3/llms.txt - [Git](https://docs.dodil.io/git) — source control: repositories over smart-HTTP/SSH, pull requests, diffs, CI checks. Full index: https://docs.dodil.io/git/llms.txt - [Registry](https://docs.dodil.io/registry) — Docker Registry v2-compatible container registry with CVE scanning and per-repo visibility. Full index: https://docs.dodil.io/registry/llms.txt - [AppID](https://docs.dodil.io/appid) — user pools for the apps you build: hosted sign-in (email/password, Google, Microsoft, OIDC), roles and permissions, tenants, service clients, and per-person MCP sign-in for Claude.ai/ChatGPT. On Ignite, attach a pool with `--user-pool` and the gateway runs the login. Full index: https://docs.dodil.io/appid/llms.txt - [Scriptum](https://docs.dodil.io/scriptum) — a typed DSL → compiled IR → durable cloud thread engine; deterministic workflows and LLM agents in one language. Inference runs on Ignite; data lives in K3. ## Notes - The unified CLI is `dodil`; products are subcommands (`dodil ignite`, `dodil data` (alias `k3`), `dodil scriptum`, `dodil git`, `dodil registry`, `dodil appid`); sign in with `dodil auth login`. Install: https://docs.dodil.io/cli - gRPC path form: `dodil..v1./`. The hub itself (this site) is product-agnostic — see each product's docs for its RPCs.