Citadail

Citadail is an AI-native hedge fund where specialized AI coworkers research companies, build stock decision pitches, pass PM/risk checks, open paper trades, and monitor the book over time.

Comment

Inspiration

What it does

Inspiration

Investment teams do not need another generic chatbot. They need a desk that can read the market, form a view, pressure-test it, create real deliverables, and keep watching after the trade is on.

The core idea behind Citadail is simple:

Imagine an investment fund or quant firm, like Citadel, made of AI coworkers. One agent finds what matters. One builds the case. One checks the risk. One places the paper trade. One keeps watch as new information comes in.

That became the product direction: not a chat app, but an AI-native investment desk.

What It Does

Citadail has two modes.

Assist Mode lets a human analyst choose a ticker, enter a thesis direction, and generate an institutional research package:

  • Investment Memo .docx
  • Operating Model .xlsx
  • PM Pitch Deck .pptx
  • PM Review
  • Risk Gate
  • Trade Desk
  • Live Book

Full Auto Mode runs a walk-forward paper investment desk from a historical start date to the present. At each simulated point in time, the system only sees data known at that moment. It scans the market, selects candidates, generates theses, runs PM/risk review, opens paper positions, monitors them, and journals every decision.

The system is paper-only. It does not claim live execution or predictive power.

How We Built It

Citadail is built around one central object: the ThesisRecord.

Every agent and artifact writes to or reads from that object:

  • ticker
  • recommendation
  • conviction
  • evidence
  • assumptions
  • catalysts
  • risks
  • invalidation triggers
  • PM decision
  • risk decision
  • paper position
  • monitoring state

The app has a web workbench for analysts and a proactive iMessage layer for PM-style updates. The Full Auto dashboard shows the simulation clock, current focus, agent activity, paper positions, book equity curve, scorecard, and audit journal.

We also built a machine-backed runtime path: Citadail can execute bounded Full Auto replay steps through an OpenClaw-compatible runtime on Dedalus, persist execution proof, and fall back locally if the runtime is unavailable.

Photon Spectrum powers the iMessage integration, letting the desk send PM-friendly updates, book summaries, ticker news, runtime status, and artifact links into a group chat.

Challenges

The hardest challenge was making the system feel powerful without feeling fake.

A normal backtest can accidentally cheat by using future data. We built the replay as a walk-forward simulation, where every source is filtered by simulationTime. Agents only see sources with timestamps at or before the current simulated date.

Another challenge was scope. It would have been easy to build a giant dashboard full of finance widgets. Instead, we focused on the actual lifecycle:

sources → thesis → artifacts → PM review → risk gate → paper trade → monitoring → audit trail

We also had to make the generated deliverables real. The memo, model, and pitch deck are actual Office files, not fake UI tabs pretending to be documents.

What We Learned

We learned that agent systems are strongest when each agent has a narrow job and a shared state object.

We also learned that investment workflows are not just about generating a good thesis. The real value is in the process around it:

  • what gets approved
  • what gets rejected
  • how much gets sized
  • when the desk adds or exits
  • what breaks the thesis
  • how the PM sees the decision

The audit trail became one of the most important parts of the product because it turns agent behavior into something reviewable.

Accomplishments

We are proud that Citadail feels like a system, not a prompt demo.

It includes:

  • a text-first analyst workbench
  • real generated DOCX/XLSX/PPTX deliverables
  • a walk-forward paper desk
  • a multi-agent investment workflow
  • a live book and scorecard
  • iMessage interaction through Photon Spectrum
  • machine-backed runtime proof through Dedalus
  • strict no-future-data replay rules
  • paper-only execution with audit logging

What's Next

Next, we would expand the historical data pipeline, add richer filing/news ingestion, improve portfolio construction, and make the iMessage PM feed more proactive.

The long-term vision is an AI investment firm simulation that can operate continuously, explain every decision, and work alongside humans instead of replacing them.

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Citadail

Built With

  • agent
  • api
  • dedalus
  • docx
  • exceljs
  • framer
  • gemini
  • github
  • imessage
  • live
  • machines
  • motion
  • next.js
  • openclaw-compatible
  • perplexity
  • photon
  • pptxgenjs
  • react
  • runtime
  • spectrum
  • spectrum-ts
  • typescript
  • vitest
Share this project:

Updates