Use GitHub with an AI Browser for Data Extraction

If you use GitHub and you regularly need to extract structured data from websites, the bottleneck is usually the same: GitHub holds part of the context, but data extraction also needs signals that live outside it - on the public web, in LinkedIn, in news, in other connected apps. Strawberry is built to combine the GitHub context with the rest of the browser, and run the full workflow as a companion you can re-trigger every week.

This page describes specifically how Strawberry handles data extraction when GitHub is one of the inputs. It names the GitHub surfaces involved, the signals the workflow actually needs, an example prompt you can paste, and what a good output looks like.

The job a researcher, ops manager, analyst, founder doing market analysis is trying to do

The goal of data extraction is to turn unstructured pages into a clean table or dataset. The success metric is concrete: extraction accuracy above 95% on spot-checked rows, dedup rate above 95%, completeness above 90%. That definition matters because it shapes what GitHub needs to contribute to the workflow.

What signals data extraction actually needs

For each signal below, here is whether GitHub can contribute directly or whether Strawberry has to find it via the browser:

• Source URL pattern (one page, paginated, search results) - GitHub does not contain this directly. Strawberry uses the browser plus public sources to fetch it.
• Target schema (which fields per row) - GitHub does not contain this directly. Strawberry uses the browser plus public sources to fetch it.
• Completion criteria (how many rows expected) - GitHub does not contain this directly. Strawberry uses the browser plus public sources to fetch it.
• Validation rules (which fields must be present) - GitHub does not contain this directly. Strawberry uses the browser plus public sources to fetch it.
• Login or paywall barriers - GitHub does not contain this directly. Strawberry uses the browser plus public sources to fetch it.
• Rate-limit posture of the target site - GitHub does not contain this directly. Strawberry uses the browser plus public sources to fetch it.

What Strawberry can do inside GitHub

Strawberry can read PR diffs, summarize issues, comment with approval, and search code across repos.

GitHub surfaces Strawberry uses for this workflow: repos, PRs, issues, commits, Actions.

How Strawberry runs data extraction with GitHub

1. Strawberry opens the GitHub repos that contains the relevant context.
2. The companion pulls related context from GitHub (PRs, history, attached files) where it exists.
3. For the parts GitHub does not store, Strawberry uses the browser - web search, LinkedIn, news, the prospect's website.
4. Strawberry synthesises the output in the shape this workflow needs: A CSV or sheet with one row per extracted entity and a confidence column.
5. A human reviews before any external action (send, update, post). Then the approved output is saved back to GitHub or your system of record.

Example Strawberry prompt

Paste this in a new Strawberry chat with GitHub connected. Adjust the specifics to your actual ICP, role, or topic.

Read this GitHub repos and any linked context.
Then run a full data extraction workflow on it. Use the browser to fill any gaps not in GitHub.
Return the output in the shape we use for data extraction: A CSV or sheet with one row per extracted entity and a confidence column.
Do not send anything externally. Save the draft to me to review.

What a good data extraction output looks like

Here is what a finished output for data extraction should look like in practice. The specifics will change for your use case, but the shape should look similar:

• Source: company directory at example.com/companies, 30 pages of 50 companies each
• Target schema: name, website, employee count, HQ city, sector tag
• Expected rows: ~1500 (50 x 30)
• Validation: name + website required; sector tag from a fixed list
• Output: ./companies.csv with 1485 rows after dedup, 12 rows flagged for human review

Why GitHub for this, and where to use a different tool

GitHub is strong for this workflow because Strawberry can read PR diffs, summarize issues, comment with approval, and search code across repos.

Where GitHub falls short Private orgs need a separate OAuth app; rate limits on large repo searches.

Consider also the rest of your stack for the parts GitHub doesn't cover.

Common mistakes when running data extraction

• No schema defined upfront, leading to inconsistent rows
• Ignoring pagination and missing 80% of the data
• Extracting from logged-in pages without confirming the cookies are valid
• Hammering the target site without rate-limiting

Connecting GitHub to Strawberry

GitHub OAuth - currently three separate apps for prod/dev/local. Once connected, the companion can read the surfaces above without re-authenticating, and any write action still requires explicit human approval the first time the workflow runs.

Caveats

Do not let any AI agent send emails, update CRM records, or change shared systems without a clear approval step. Strawberry is strongest when the workflow combines browser context with connected-app context and a human review for sensitive actions.