Auditable AI investment intelligence
Investment research.
Accessible. Inspectable.
Accountable.
Aegis aims to make disciplined investment intelligence accessible beyond institutions: read the world, form a testable view, measure the decision, retain what was learned.
The goal is a useful investment engine people can understand. The system is in development; its research is evaluated through recorded experiments and paper accounts.
Architecture motion study. Stage animation explains the system; it is not live worker telemetry.
Historical paper highlights · 09 October 2026
Measured work.
A result with its record.
The selected historical examples below include the original +10.599% simulator observation. They are dated highlights, not current returns or a complete account history.
| Experiment | Window | Mode | Return | SPY |
|---|---|---|---|---|
| Revision flow | 2026-09-28 – 2026-10-09 | Simulator | +10.599% | +1.330% |
| Net raises | 2026-09-28 – 2026-10-09 | Simulator | +9.759% | +1.330% |
| Hack2 | 2026-08-28 – 2026-10-09 | Broker paper | +5.517% | +1.219% |
Three positive examples selected retrospectively from 307 benchmark-comparable accounts. Different windows are not one portfolio. Simulator results and broker-paper observations are separate; Hack2 has actual series for 10/30 sessions. These observations do not establish repeatable investment skill. Dates, selection and source record.
One loop · every belief graded
From the world
to a measured decision.
Language models help interpret information. Explicit code ranks, sizes and grades decisions. A useful research system needs both the reasoning and the record.
World sensors
Collect market information, filings and analyst revisions while retaining source identity and capture state.
news_pull.py · web_reader.py · pull_analyst_targets.pyDevelopment record
More than a dashboard.
The implementation joins market research, quantitative experiments, paper evaluation and persistent project memory. Its history begins in March 2026. The public repository is a curated account of that work; the original operational implementation is private.
Research collection, evidence processing and decision support are potential services in their own right. They share a practical foundation: source acquisition, reproducible processing, explicit uncertainty and a record that another session can inspect.
- Research & models
- Hypotheses, deterministic decision contracts, benchmark comparisons and paper-account experiments.
- Learning from work
- Pre-registration, source attribution, negative findings and branch/review records inform subsequent experiments. They do not automatically certify an investment edge.
- Persistent context
- Optimus retains selected source-linked knowledge and reusable instructions, available to compatible assistants through MCP.
Visual library
The system, explained.
Original orbit and blackline diagrams. Vector motion follows a stage, a source connection or a learning path.
Orbit / research loop
Blackline / experiment path
Application prototype
A working interface.
An evolving system.
The application interface has its own deployment. This showcase explains the project and its record; the app lets visitors explore the interface work.
Backend hosting is paused to control server costs. The application link is an interface preview; do not expect live analysis or current account data. No fabricated live results are inserted.
Make intelligence useful.
Keep the reasoning inspectable.
Investment research should be easier to access and harder to misrepresent. That is the direction of Aegis.