App - Hindsight

Overview

Backtesting an investment idea normally means notebooks, data wrangling, and a lot of glue code before you learn anything. Hindsight collapses that into a conversation.

You write something like "Short every stock Jim Cramer recommends, $10,000 start, 2016 to 2025." An agent writes the strategy code, fetches the historical data it needs, runs the backtest, and streams each step as it works. What comes back is a single equity curve — green if the idea worked, red if it didn't — plus a chat box to refine the rules and run it again.

Hindsight is local and single-user by design. Every strategy lives in its own session on disk, so the code, the chat, and the results are all exactly where you left them after a restart.

Historical data only. Nothing is traded.

Key features

• Plain-English strategies Describe the idea and the period in a sentence. No notebooks, no data pipeline to build first.

• Watch the agent work Runs stream over Server-Sent Events, so you see the steps — strategy written, prices fetched, backtest running — instead of staring at a spinner.

• Refine by chatting Tell it what to change and the agent edits the existing strategy. Changing just the dates re-runs the saved code with no LLM call at all.

• Sessions persist One session per strategy, stored as JSON on disk. Close the app, restart the server, come back next week — nothing is lost.

• Two agent backends A built-in mock agent gives a fully usable, deterministic demo with no Docker or API keys. Switch to the Codex backend and each run gets its own Docker container that writes real Python and runs a realistic backtest.

• Container-isolated execution Real runs execute inside a per-session container with an isolated filesystem, torn down afterwards — the sandbox is the container, not a permissions prompt.

Open source

Hindsight is MIT licensed and self-hosted. Bring your own agent and your own market data. The source lives at CleveroAB/hindsight, and the project site is at hindsight.build.

Tech & focus areas