Inalpha

Find alpha with a fox's eye.

The quant familiar · backtest = paper = live

Conversational AI agents for quantitative research, strategy authoring, and paper trading — wired with Claude Code–grade engineering discipline.

View on GitHubRead the docs

The loop

One loop. User → agents → kernels → strategy.

An orchestrator routes your prompt across three Python kernels. Decisions flow back into the conversation — fully transparent.

YouOrchestratordatapaperresearchStrategy

Three kernels

Three Python services, one conversation.

data/ python

Multi-venue feeds: crypto, US, A-share, HK, JP, KR, AU, IN, UK, DE equities, global indices, FRED macro.

from inalpha_data import get_bars
paper/ python

In-memory matching, backtest engine, and persistent paper trading with replay-able state.

from inalpha_paper import run_backtest
research/ python

Multi-analyst LLM debate. Freshness-anchored. No stale numbers passed off as insight.

from inalpha_research import debate

Why Inalpha

Engineering discipline, not vibes.

  • 01

    Unified kernel

    One strategy codebase across backtest, paper, and live. Behavior must be identical — or nothing is meaningful.

  • 02

    Agents are first-class

    Research, decision, risk, and review have dedicated agents — opposing stances, distinct toolsets, traceable decisions.

  • 03

    Transparency over precision

    Prefer an agent that says “I don't know” over one that sounds certain but cannot show its evidence.

  • 04

    Discipline over shortcuts

    Decision records, tests, declarative guards come first. Clever code is a bug nursery.

  • 05

    Long-horizon compounding

    Solid infrastructure before flashy features. Surviving long matters more than running fast.

Coverage

Markets we already cover.

Crypto-first, multi-venue from day one. Each market routes through the same kernel and the same agent prompts.

  • Crypto
  • US Equities
  • A-share
  • Hong Kong
  • Japan
  • Korea
  • Australia
  • India
  • UK
  • Germany
  • Global Indices
  • FRED Macro

Built different

Claude Code–grade engineering harness.

Declarative, config-driven, and reviewable. Not a chat wrapper.

hooks

Middleware on every tool call.

permissions

Role-scoped tool access.

plan-exec

Plan-then-execute with one-shot tokens.

subagent

Isolated subagents for risk and review.

MCP

Model Context Protocol native.

Swarm

Parallel workers for grid backtests.

Get started

Star it. Read it. Tear it apart.

Inalpha is alpha-stage and AGPL-3.0. No real money yet — but every line is on GitHub.

$git clone https://github.com/mirror29/inalpha
View on GitHub