Skip to main content

AI code research. Understand the system.

Research a repository, compare libraries or plan a migration. Follow the explanation back to source files and documentation.

Try
OpenClaw · public DeepDive

How does OpenClaw coordinate chat, tools and sessions?

The investigation follows a message from the browser console into the gateway and agent runtime. A shared gateway coordinates identity, session routing and tool access across interfaces. The report maps the user journeys, explains plugin and memory boundaries, and identifies operational questions to review before wider adoption. Read the full DeepDive to inspect the architecture, assessment and supporting implementation details.

Read full report
nanobot · public DeepDive

How does nanobot keep a tool-using assistant compact?

This report examines a small assistant framework with a tool-calling loop, workspace memory and a gateway for chat channels. It follows how incoming messages become a session, how context is rebuilt and how older conversations are consolidated. The architecture discussion connects the shared message bus and channel adapters to the goal of keeping platform-specific behavior outside the core agent.

Read full report

How it works

Try it now
  1. 01

    Say what you want

    A sentence or two is enough.

  2. 02

    Work with HowWorks Agent

    Describe the result you want. The Agent works with the tools you choose.

  3. 03

    Make it

    Start creating, then keep working with HowWorks.

About

The useful question is often specific: where does this framework store a document, which component handles retries, or what would a database change affect? AI Code Research helps you investigate the repository around that question. For open-source projects, ask for the files and code paths behind the explanation. For closed-source tools, keep the investigation grounded in accessible documentation, SDKs and public issues. Use the findings to choose what to read, test or build next.

Questions

What is AI Code Research?

AI Code Research is a focused investigation of a codebase or technical choice. Use it to explain architecture, trace a workflow, compare libraries or examine a migration. The request should identify the project, the question and the evidence you want in the answer.

What is a good first question about an unfamiliar repository?

Start with one behavior you need to understand: how a document is saved, how an API request is authenticated, or what happens after a failed job. Ask for the entry point, the relevant modules and references to the source files. Expand from that path into the wider architecture.

Can I compare two libraries or frameworks?

Yes. Specify the workload and constraints, such as deployment options, persistence, extensibility or upgrade costs. Ask the agent to compare those requirements against accessible source and documentation, and to separate observed behavior from assumptions that still need a test.

Can it explain a closed-source tool?

The investigation is limited to accessible evidence, such as official documentation, public SDK code and issues. It can explain documented interfaces and reported behavior. It should not claim to have read private source code or verified an internal architecture that is not public.

Will a research report prove that a migration is safe?

A report can identify dependencies, compatibility questions and a proposed sequence of changes. The migration still needs tests and review against your actual application and data. Ask the report to make those remaining checks explicit.

From the blog

Share image · Milky Way Core · ESO/VVV Survey/D. Minniti · Acknowledgement: Ignacio Toledo, Martin Kornmesser · CC BY 4.0 · Cropped