Let new evidence challenge your investment thesis.
Who it is for
Section titled “Who it is for”AgenticThesis is for self-directed, long-term investors and developers who want a thesis record that can survive new evidence. You state why a business deserves attention and what would prove each reason wrong; the engine monitors selected official sources, proposes evidence-linked updates, and preserves the review trail.
It is also a reusable backend engine. The same Python distribution exposes the AgenticThesisEngine interface and the agentic-thesis serve self-hosted application.
Visit the GetSuvi official website for future updates to Suvi’s broader investment-agent product family. AgenticThesis is its open-source thesis-monitoring engine.
What it owns
Section titled “What it owns”- discovery of configured SEC and issuer IR events;
- immutable source artifacts with hashes and parser outcomes;
- thesis-aware Radar routing before model use;
- source-addressable evidence and exact citation validation;
- recoverable runs, Human Review, and immutable thesis revisions.
AgenticThesis does not value securities, size positions, predict prices, or issue Buy/Sell/Hold instructions. The investor owns the final judgment.
A concrete example
Section titled “A concrete example”An earlier measured Apple run compared three beliefs with a new filing:
| Saved belief | New company evidence | Proposed result |
|---|---|---|
| Services improve Apple’s business economics | Services gross margin remained materially above Products and Services revenue grew | Supported |
| Greater China demand remains steady | Greater China sales declined, driven mainly by lower iPhone and iPad sales | Possibly invalidated |
| Supplier concentration will not stop production | Concentration remained disclosed, without evidence that production had stopped | Weakened |
Each result retained exact source evidence. None became thesis history until approval. This is an illustrative historical run, not investment advice or a claim about current Apple fundamentals.
System shape
Section titled “System shape”
One application-owned LangGraph coordinates the state transitions. Deterministic code owns parsing, retrieval fusion, citation validation, Radar routing, and version commits; model work is limited to embeddings, conditional reranking, and structured thesis comparison.