Agentic Quant Studio

Agentic Quant Studio

A workspace for building agentic AI systems in quantitative finance.

---

What exists today

This repo is an early-stage data platform, not yet an agentic workspace. There is no chat, no RAG, and no backtesting engine in the codebase. What works end-to-end today:

• Warehouse — Hive-partitioned Parquet, catalog scan, read/resample ([crates/warehouse/README.md](crates/warehouse/README.md))
• Backend — Axum API, jobs, catalogs, studies ([crates/backend/README.md](crates/backend/README.md))
• Frontend — Market Research + Data Management ([frontend/README.md](frontend/README.md))
• Studio — GraphSpec + runtime + indicator catalog ([crates/studio/README.md](crates/studio/README.md))
• MCP server — stdio tools, resources (docs/schema/examples), and prompts for external agents over the backend API ([crates/mcp-server/README.md](crates/mcp-server/README.md))

The name reflects the long-term vision (see [Vision](#vision)); the implementation is focused on reliable market data, chart UX, and the graph spec foundation for indicators and strategies.

MCP integration path

The MCP server (aqs-mcp) exposes the backend (catalogs, graph validation, draft studies) over the Model Context Protocol, plus resources (aqs://docs/…, aqs://schema/…) for GraphSpec documentation and node-kind discovery, and prompts for common draft workflows. Host apps and external agents (Claude, Cursor, Grok, or anything else that speaks MCP) can drive the platform today—before an in-product agent exists—and remain a first-class alternative path once built-in agents land.

Demo: MCP walkthrough on YouTube — external agent → GraphSpec draft → accept in Market Research.

Other tools and data sources can meet AQS at the same API surface. Agent-authored studies land as drafts the user accepts in the Market Research UI. Setup and tools: [crates/mcp-server/README.md](crates/mcp-server/README.md).

---

Current features

Frontend

Two pages: Market Research (/) — LW Charts candle view with catalog-driven indicators; Data Management (/data) — ingest, datasets, jobs.

The chart uses a layer document → compiled ChartBlockSpec → studio run pipeline. Indicator metadata comes from GET /catalog/indicators; the UI hydrates a runtime registry on load.

Details — architecture, components, datafeed, tests, dev setup: [frontend/README.md](frontend/README.md)

---

Backend & data

HTTP API at /api/v1 (candles, catalogs, jobs, studio runs). Ingestion uses an in-process job queue; candles live in local Parquet (Bybit only today).

Details — endpoints, curl examples, jobs, config: [crates/backend/README.md](crates/backend/README.md)

Warehouse layout, catalog, resampling: [crates/warehouse/README.md](crates/warehouse/README.md)

Studio

Declarative GraphSpec + validate/execute runtime. Powers POST /studio/runs and GET /catalog/indicators (via IndicatorCatalog + chart_defaults).

Details — spec types, built-in nodes, examples: [crates/studio/README.md](crates/studio/README.md)

---

Tech stack

| Layer | Stack | |-------|--------| | Backend | Rust, Axum, Tokio | | Studio | Computation graph spec (GraphSpec); runtime WIP | | Jobs | In-process queue + worker (not Redis/Sidekiq) | | Warehouse | Parquet, Polars, custom catalog | | Frontend | Next.js 16, React Query, Zustand, shadcn/ui, Lightweight Charts, Vitest | | Agents / hosts | MCP server (aqs-mcp) over backend HTTP; in-product agents later |

Not in the repo yet: in-product agent/chat (e.g. Rig), RAG, backtesting, MLOps.

---

Getting started

git clone https://github.com/wizard50/agentic-quant-studio.git
cd agentic-quant-studio

Backend

cargo run -p backend

Config and API examples: [crates/backend/README.md](crates/backend/README.md)

Frontend

[frontend/README.md](frontend/README.md) — dev server (use a different port than the backend), chart architecture, npm test.

---

Project structure

/
├── config/              # defaults.toml, example.toml
├── crates/
│   ├── api-client/      # exchange clients (Bybit)
│   ├── backend/         # Axum API — see crates/backend/README.md
│   ├── common/          # shared types
│   ├── mcp-server/      # stdio MCP path — see crates/mcp-server/README.md
│   ├── studio/          # GraphSpec + runtime — see crates/studio/README.md
│   └── warehouse/       # Parquet + catalog — see crates/warehouse/README.md
├── frontend/            # Next.js UI — see frontend/README.md
└── README.md

---

Vision

Long-term goal: an intelligent workspace where users interact with AI agents (chat, later voice) to:

• Run quantitative research and backtesting
• Generate indicators, strategies, and dashboards
• Use RAG on documents and private knowledge bases

None of that is implemented yet as a first-party agent experience. The MCP server is already the open path for external agents and host apps, and is expected to stay as an alternative even alongside built-in agents (see [MCP integration path](#mcp-integration-path)). The current milestone remains reliable market data ingest + catalogs + chart UX with dynamic indicators + GraphSpec foundation for agent-composed strategies.

---

License

This project is licensed under the [MIT License](LICENSE).

---

Built by @wizard50

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: wizard50
• Source: wizard50/agentic-quant-studio
• License: MIT
• Homepage: https://www.youtube.com/watch?v=V80_Kx98FMM

Install and usage instructions live in the source repository linked above.