An open-core agentic stack.

Modular by design.

Ten composable modules — orchestration, retrieval, quant research, chat, auth, observability. Self-hosted from day one.

Tour the stack GitHub

Act I · Orchestration

DigiGraph

The brain. A LangGraph supervisor routes every request.

A declarative tool registry pairs with a sub-graph supervisor. Any vertical exposes capabilities via POST /v1/orchestrator_tools and drops straight in.

LiteLLM routes calls; LangGraph checkpoints keep state durable across hops. Speaks OpenAI API on the front door.

Act II · Quantitative

DigiQuant

The flagship vertical. Research, execute, audit.

NautilusTrader core with a strategy registry. Atlas reasons over research, Hermes publishes signals, Kairos executes — loopback-only, human-gated.

Open digiquant.io

Act III · Retrieval

DigiSearch

Ingest. Chunk. Embed. Search. Polars throughout.

One DigiSearch client, pluggable backends — Chroma, pgvector, Qdrant, in-memory. Backend-neutral entities: Document, Chunk, Query, Result.

Multi-mode retrieval (dense / sparse / hybrid) selected per query.

Act IV · Chat

DigiChat

Next.js chat UI + BFF. Streams DigiGraph. BYOK.

Machine keys or user session — DigiChat brokers DigiGraph traffic with the right bearer, tenant, and proxy key. Adapts to whatever scopes the JWT grants.

Open DigiChat

Act V · Ecosystem

Plug in. Expose. Ship.

The same diagram, three deployment stories. Pick a lens.

Each lens overlays the diagram.

Module

…

Act I

DigiGraph

Route any request through the right specialist agent.

Act II

DigiQuant

Backtest, optimize, and deploy trading strategies.

Act III

DigiSearch

Find the right document. Ground any answer.

Act IV

DigiChat

Talk to your stack. Ship with your own keys.

Act V

Ecosystem

Plug it in. Expose as MCP. Ship as Docker.