Workspace / Overview
Simulated fundsModel not connected
Your THESIS workspace

Explore saved AI reviews, citations, paper accounts and the interactive demo lab. These records retain their original dates. Current news and quotes for all 19 stocks are available in Announcement review, with no API key.

Your decision workspace

Every trade needs a thesis.

Follow the news. Check what the price has already absorbed.
Make the reason for every action visible.

AI decisions recorded
0
Observed run · excludes demo
Simulated fills
0
No real money moves
Companies in saved run
02
Apple · NVIDIA
AI connection
Not connected
Connect a model to begin

From an event to an action

The path each eligible event follows.

How it works
1

Read the original announcement

Collect an issuer's news feed and preserve its source and timing.

2

Ask what is already in the price

The model considers the announcement, the observed price response, and missing information.

3

Check the proposed trade

Independent rules inspect quote freshness, trading costs, position size, and liquidity.

4

Act and keep the evidence

Record a simulated fill, a rejection, or a decision to wait.

Run status

A clear view of what is actually connected.

Paused
Official news

Saved run: Apple · NVIDIA. Read current news for all 19 stocks →

Not collected
Bitget market data

Quotes and instrument information

Not collected
AI decision maker

Needs a model connection

Setup needed
Ready to collect observations.

The runner works while the local app is open. Pause stops monitoring; an active request may finish. Model calls may incur provider charges.

Agent & comparison rule

Separate portfolios with matching execution limits. No advantage has been demonstrated.

Original sources · observed prices

The inputs behind a decision.

Quote availability, publication times, and trading-session checks remain visible.

Collect observations to see the available quotes.

Issuer announcements

Only new, eligible events enter a paper cycle. Older items remain visible for context.

0 events
No announcements collected yet.Choose “Refresh inputs” to contact the official news feeds.
Original announcement · Source evidence

What does the news actually change?

Read the source, assess what it changes, and choose the next evidence to investigate.

Research only

Research connection

Using the existing model.

Research only
Set up Groq for source briefs and assessments
The key stays in this app session and is cleared from this field after submission. Re-enter it after restarting THESIS.

Groq receives the collected public issuer passages when you request a source brief or assessment. The paper decision model is separate. Keep your Groq account on the Free plan; THESIS cannot check or change billing.

Choose an announcement

Older news can be reviewed here. Its original publication time stays visible.

Step 1: review the source. Step 2: use Assess evidence to ask what it changes, what could weaken the idea, and what to check next. Each step makes one model call. Research does not place trades.
Evidence before action

A useful answer can be “not enough evidence.”

Choose an announcement and start a review. You will see what supports the idea, what challenges it, and what the model cannot conclude.

01   Read the source02   Check the price response03   Challenge the conclusion
Simulated portfolios · Recorded decisions

See what THESIS did, and why.

Follow positions, costs and the evidence behind each outcome. Each instrument and decision maker has a separate account.

Decision journal

Latest 20 records per account. The export contains the complete history. Research reviews and scripted demos are separate.

Understand the decision process

A trade is only one possible answer.

Explore six situations through the same execution checks used by the paper engine.

Synthetic examples
This lab uses scripted decisions and invented market inputs. It demonstrates software behavior, not AI reasoning or trading performance.
Loading demonstration…
Connect the decision maker

Give THESIS its AI connection.

Use a model that supports JSON responses through a Chat Completions endpoint, such as Qwen.

Paper decision model

The key stays in this local app's memory and is sent only to the endpoint you provide.

Online providers must use HTTPS. A local model may use http://127.0.0.1:11434/v1/chat/completions.
It is not saved to a file or browser storage. Restarting the app clears it.

This sends a small connection-test request. An online provider may charge; a local model uses your own computer.

Start in three steps

  1. Launch the downloaded project with python3 launch.py and open the local address it prints.
  2. Connect an online model here, or run an OpenAI-compatible local model first and use its loopback endpoint.
  3. Refresh the inputs. Once the data is eligible, run one paper cycle and inspect the result before starting continuous monitoring.
No Bitget trading key is needed for the local simulator. Local-model HTTP is accepted only on your own computer; every other provider must use HTTPS.

What remains to prove

The agent still needs genuine model-backed forward logs and a measured comparison against a simple rule. A working interface is not evidence of investment performance.

THESIS walkthrough

The video demonstrates the local app. This browser preview uses saved records.