Description
Open ETF Agent is an open-source project designed to provide a practical, verifiable framework for building and understanding ETF research agents. It functions as a runnable field guide, enabling users to move from theoretical articles to executable modules. The core philosophy revolves around the ability to read about a mechanism, run its corresponding implementation, inspect the generated reports, and trace back to the underlying code, original sources, and execution evidence.
Read More
The project emphasizes transparency and verifiability, allowing users to understand how an agent works by observing its output and tracing its development history. It offers a series of modules, each representing an evolutionary step in the agent's development, rather than just article publication dates. These modules demonstrate specific functionalities, such as persistent human-agent interaction, temporal awareness, and PDF content extraction.
Users can try the agent locally, requiring Node.js and npm. The default examples run offline and do not require model keys, making it accessible for experimentation. The command-line interface generates reports, manifests, and run records, providing a clear audit trail of execution. This setup is designed for synthetic examples, not real-time fund research, focusing on the mechanics of agent operation.
The project also maintains a detailed build log, documenting incremental improvements and fixed development snapshots. Each stage highlights a single, independent delivery focus, allowing users to examine the original code at any point. This includes a comprehensive version history, from the first runnable report (v0.1) to more advanced features like content trust and persistent human-agent interaction (v0.13).
A significant aspect of Open ETF Agent is its focus on content trust and security, particularly concerning prompt injection. It implements mechanisms to establish content trust boundaries, ensuring that external content is treated as data rather than instructions that can alter system behavior. This involves preserving original identities, verifying actual SDK projections, and preventing content from self-attributing authority. The project explicitly separates source and evidence identity from instruction identity and execution authorization, aiming for a robust and verifiable system.
The project also delves into complex topics like cross-run conversation compaction, addressing how an agent manages historical context without compromising the integrity of original facts or user intent. It outlines a conservative strategy for summarizing old assistant responses while strictly preserving user prompts and source identities, ensuring that summaries do not overwrite original facts. This meticulous approach to data handling and execution tracing makes Open ETF Agent a valuable resource for developers and researchers interested in the practical engineering of AI agents.
- CategoryAI Agent Builders
- Pricing ModelOpen Source
- DeploymentLocal
- PlatformsNode.js
Open ETF Agent's Core Features
Runnable Field Guide: Execute modules to understand agent architectures.
Version Tracking: Trace code, versions, and evidence boundaries.
Open Source Implementation: Access and inspect the underlying code.
Synthetic Examples: Run default demonstrations without model keys.
Local Execution: Try the agent on your local machine with Node.js.
Report Generation: Produce readable reports, manifests, and run records.
Detailed Build Log: Review incremental development snapshots and version history.
Content Trust Boundaries: Implement A35 content safety against prompt injection.
Persistent Human-Agent Interaction: Manage durable user conversations.
Temporal Awareness: Handle task dates and IANA time zones.
PDF Content Extraction: Process local text layers from PDF documents.
Context Compaction: Summarize old assistant responses without altering original facts.
Modular Design: Explore agent evolution through distinct modules.
Verifiable Execution: Ensure every claim is traceable to code and evidence.
GitHub Releases: Access stable versions and development history.
How to use Open ETF Agent?
Install Node.js: Ensure Node.js 26.8.1–26.x and npm are installed.
Clone Repository: Obtain the Open ETF Agent source code from GitHub.
Install Dependencies: Run `npm install` to set up project dependencies.
Run Locally: Execute CLI commands to generate reports and records.
Read Reports: Open `report.md` in the `runs/` directory to inspect outputs.
Inspect Code: Review the source code for each module and version.
Explore Modules: Follow the reading map to understand agent mechanisms and boundaries.
Open ETF Agent's Use Cases
- Agent Architecture Study
- Verifiable AI Development
- Prompt Injection Defense
- Context Management in Agents
- Local AI Agent Prototyping
- PDF Content Processing
- Durable Human-Agent Interaction
- Open-Source Project Contribution





