THE SOURCE / A LAB BY KEITH DE ALWIS

Open-source intelligence for people and AI agents.

Discover software. Understand project health and momentum. Find credible open-source alternatives, with evidence you can inspect.

9,364open-source projects discovered
235verified and scored in depth
309recommendations, each with its evidence
SCROLL TO FOCUSDATA AS OF · DATASET 02138adc
01 / SURGE RADARUPDATED

What is moving right now.

Star velocity across the 625 projects The Source measures every day, and the repositories developers are talking about. Every figure is measured, dated and explained.

7-DAY WINDOW00PROJECTS ON RADAR

    02 / THE QUESTIONSWHAT IT IS BUILT TO ANSWER

    Ask it what you would ask a good analyst.

    1. 01

      “What are credible open-source alternatives to Notion, and what are the trade-offs?”

      source_get_alternatives
    2. 02

      “Is this project actively maintained? Show the signals and their dates.”

      source_get_project
    3. 03

      “Which self-hostable alternatives to Slack are actively maintained?”

      source_recommend
    4. 04

      “Compare two projects for a 50-person team. What should we verify before adopting?”

      source_compare_projects
    03 / TWO WAYS INONE INTELLIGENCE LAYER

    One intelligence layer. Two ways in.

    FOR HUMANS

    Explore the ecosystem.

    For founders, developers, CTOs and researchers. Discover projects by category or by the commercial product they replace, compare the options, and judge fit on more than star counts.

    • Discover by category, use case or commercial alternative
    • Read health signals, release activity and momentum, with dates
    • Compare candidates with trade-offs and source links
    Explore alternatives
    FOR AGENTS

    Give your agent a source.

    A Markdown onboarding file your agent can read for itself. It learns what The Source is, picks the connection route for its client, asks you for any approval it needs, confirms the tools, then runs a sample query.

    • Machine-readable onboarding at agents.md
    • Public, read-only MCP server. No account
    • Every answer carries its date, provenance and licence
    Point your agent at The Source
    04 / THE INTERFACELIVE FROM THE PUBLIC READ LAYER

    Evidence, not opinion.

    Pick a product. These are the real, ranked recommendations from the published library, fetched live as you choose. What was observed is kept apart from what was inferred.

    api/v1/saas/notion.json
    REPLACING

    Notion

    Knowledge management
    ObservedInterpretation

    Loading the published snapshot…

    Replacement Fit and Project Health are machine-derived scores under methodology v1.2.0, not editorial verdicts. Stars, licences, releases and activity dates are observed from each project’s own host. Verify before you adopt.

    05 / OPEN BY DESIGNINSPECTABLE AND REUSABLE

    The intelligence is yours to inspect.

    9,364projects in the discovery tier. Searchable, labelled, never presented as recommendations
    235projects verified and scored for fit and health
    117commercial products mapped to open-source replacements
    9upstream sources, each licence-reviewed and pinned to a commit
    CODEMIT
    DATACC BY 4.0
    ACCESSGitHub + MCP

    The repository licence does not automatically cover every upstream dataset. Source-level terms are documented separately, parts adapted from a share-alike source carry CC BY-SA 3.0, and 25 relationships are withheld from export because of them. Read the data licence ↗

    06 / HOW IT WORKSA DAILY INTELLIGENCE CYCLE

    Five steps. Every day.

    1. 01

      Harvest

      Nine licence-reviewed upstream lists, each pinned to a reviewed commit, plus a manual intake ledger. A list only advances if its licence file is unchanged.

    2. 02

      Verify

      Every candidate is checked against its own repository host. Maintained in the last twelve months, under a recognised open-source licence, or it does not pass.

    3. 03

      Score

      Replacement Fit and Project Health, each with a confidence figure and a versioned methodology. Machine-derived, and labelled as such.

    4. 04

      Publish

      A daily cycle refreshes the evidence and writes a dated, hashed snapshot. The read layer is static JSON you can cache, diff or audit.

    5. 05

      Serve

      The same layer, through a public read-only MCP server and the GitHub library. One source of truth. Nothing downstream keeps its own copy.

    07 / AGENT ONBOARDINGREAD → CONNECT → VERIFY → ASK

    Point your agent at The Source.

    01

    Read

    The agent reads agents.md: what The Source is, what it can answer, and what it cannot.

    02

    Connect

    It picks the route for its client. Some clients ask you to approve a connector or edit a local config. That is a feature.

    03

    Verify

    It lists the tools and checks the dataset version and freshness before trusting an answer.

    04

    Ask

    It runs a sample comparison, and cites the dates and sources behind the result.

    THE FASTEST ROUTE · ANY AGENT THAT CAN READ THE WEB

    Read https://de-alwis.com/lab/the-source/agents.md, follow the setup for this client, confirm the available tools, then compare two open-source alternatives to Slack with evidence.

    CONNECT CLAUDE CODE
    1. Run the command once. It adds The Source to your Claude Code configuration.
    2. Start Claude Code and type /mcp to confirm the-source is connected.
    3. Ask your question. Claude Code calls the tools it needs.
    Run in your terminal
    claude mcp add --transport http the-source https://source-mcp.de-alwis.com/mcp
    Then ask
    Compare two credible open-source alternatives to Slack. Cite the dates and sources behind each figure.
    THE PROMISE

    Less searching. Less guesswork. Better software decisions.

    A CONNECTION STARTS HERE

    Let’s talk.

    An idea taking shape. A challenge worth solving.
    Tell me what’s on your mind.

    Your details are used to respond to this enquiry.