EvoQuant CLI
A CLI agent for quantitative investment research: it reads broker reports, ideates factors and computes IC on offline data.
- 142 ★ on the repository
- Last commit
Extended access
Why this level
- The agent executes generated experiment code and calls the chosen model provider's API under your key.
- Telegram, Slack and other channels plus scheduled runs work unattended and forward results outward.
Install
Manual install
uv tool install "git+https://github.com/CamusGIT/EvoQuant"Python 3.11 to 3.13 and `uv` are required. Launch `evoquant` after installing and put the provider key into `.env`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent executes generated experiment code and calls the chosen model provider's API under your key.
- Telegram, Slack and other channels plus scheduled runs work unattended and forward results outward.
This is third-party code. Review the repository files before installing.
What it does
EvoQuant is a standalone terminal agent that runs a quant analyst's research loop. It turns broker PDF reports into a structured knowledge base, searches a local paper library, generates ideas and ranks them in an ELO tournament. The agent then rebuilds the described method as an executable artifact and tests it on offline data with IC, ICIR and RANKIC metrics. Inside are a team of six sub-agents (plan, research, code, debug, analyse, write), an observation memory and 20 skills in `EvoQuant/skills/`. The agent works in the CLI and TUI, in a web UI (`--ui webui`) and in Telegram, Slack, Feishu and Discord channels, and it supports MCP and scheduled runs.
Who it is for. Quant analysts and factor researchers who want an AI assistant for reproducing methods from broker reports that ship without open code.
Good fit when
- A folder of broker PDF reports must become a searchable knowledge base with citations.
- A method from a report must be reproduced from scratch and tested on offline data.
- Research ideas are ranked by novelty, relevance and clarity minus difficulty.
- Research cycles run unattended on a schedule with reports sent to Telegram or Slack.
Not a fit when
- You need trading signals and order execution: a backtest-to-live bridge is only on the roadmap.
- You need a plug-in skill for Claude Code: EvoQuant runs as a separate agent.
- You have no API access to any supported model provider.
Example request
Find papers about cross-sectional momentum and propose a research direction in alpha factors.Limitations
Python 3.11 to 3.13 (`uv` recommended) and a key from one provider are required: Anthropic, OpenAI, Google, MiniMax, NVIDIA or OpenRouter. The key goes into `.env` created from `.env.example`. PDF reports are placed in `papers/raw/`. The agent searches papers only in the local library, and web search is not used for them. Quant metrics are computed on offline data that must be prepared in advance. Portfolio metrics, alpha-generation methodology and a bridge to live trading are marked as unfinished on the roadmap. The repository's last commit is dated September 2026.
How to disable. Remove the installed tool with `uv tool uninstall evoquant`, stop the container if you ran Docker and delete the `.env` with keys. Clean the knowledge base in `papers/` separately.
FAQ
How do I install EvoQuant?
Run `uv tool install "git+https://github.com/CamusGIT/EvoQuant"` and launch `evoquant`. Another route: clone the repository, run `uv sync` and `uv run evoquant`. Docker works too: `docker compose up`.
Which metrics does the agent compute?
IC, ICIR, RANKIC and coverage. The calculation runs on offline data that the runtime discovers itself, then builds a panel and splits it into train, validation and test.
How does the idea tournament work?
An idea inherits at least 70% of an anchor report's method and adds up to 30% novelty. The final score is novelty plus relevance plus clarity minus difficulty, and the winners go to experiment.
How do I add more skills?
With `/install-skill` inside a session or through the skill browser. The agent also installs skills from GitHub on request.
Is there a web interface?
Yes, launching with the `--ui webui` flag opens a workspace panel in the browser. The CLI and TUI mode stays the main one.
Discussion
No questions or reviews yet.
Related
CLI toolsThe MCP server, CLI and SDKs of the QVeris platform: the agent finds a data service, calls it and audits credit spending
SearchCLI is a Volcengine AI Search tool that evaluates results against test queries and retains an experiment plan and report.
A skill and Python package for hyperparameter tuning: the agent reads the code, proposes the next set of values, and your own run measures the result
A Claude Code and Codex plugin that builds an ontology over your tables and files and refines it from the agent's work history
A Claude Code plugin and MCP server: over 1000 scientific tools for biology, chemistry and medicine, data-check commands and routing skills
MCP server that connects Claude Code and Codex to an open R or Python session in RStudio and Positron, where your data lives
README in short
The project rests on the idea that quant methodology lives in broker reports without open code. The agent reproduces a method from its description and checks it on real offline data instead of trusting the numbers printed in the PDF. The base is DeepAgents and LangGraph.