Start with content, then trace its sources
Creator Source Graph v0.7.0 starts with a product or project URL. Your AI derives 6–10 relevant keyword angles, then searches for related and similar content on exactly four social platforms: Instagram, YouTube, X and Threads. It opens originals and records view counts where they are visible; unavailable counts stay unknown.
From those posts, it follows citations and links through intermediate articles, projects and papers, up to six hops. Where links leave gaps, the web app’s agent compares distinctive claims and examples with opened candidate sources. Supported candidates appear as separate inferred connections; missing evidence stays unresolved. The graph then helps evaluate where new information about your product could be published or listed.
Use the AI host you already have
The initial installed skill can continue researching in the same Codex or Claude Code chat that installed the app. Analyze and Reanalyze in the browser start a new local job through an installed, signed-in Codex or Claude Code CLI. They do not resume the chat you already have open.
Auto chooses an available subscription host, or you can select one. Before replacing the current analysis, the app checks that the host is compatible and signed in. If that check fails, it keeps the previous graph and explains whether to install, update or sign in. Your existing plan, web tools and usage limits apply; no provider API key setup is required.
Give your AI the GitHub URL
Download Creator Source Graph for macOS Apple Silicon, macOS Intel, Windows x64, or Linux x64. These packages include Node.js. A portable source ZIP is also available for users with Node.js 20 or newer.
Give your existing local AI host this instruction: https://github.com/agentlas-ai/creator-source-graph — Install and open this app, then research creators covering https://agentlas.cloud. Replace the final URL with your target. The host reads the setup instructions, obtains the app, installs its skill and opens Agentlas sign-in. The installation repository URL does not automatically become the research target.
You can also open the extracted package with Start.command on macOS, Start.bat on Windows, or start.sh on Linux. The package includes skill installers for both hosts. In a later Claude Code chat use /creator-source-graph <target-url>; in Codex use $creator-source-graph <target-url>.
Sign in, enter a URL, then analyze
Agentlas sign-in leads to a separate home screen with a centered URL input. Enter your target and choose Analyze to open the analysis view. Use Back to return home and enter a new URL, or View latest analysis to reopen the latest investigation. If you supplied a target during installation, the initial skill can continue with that target after sign-in.
In the analysis view, edit the target and reanalyze. The screen keeps the research compact: icons, names, one short summary and original links. Detailed metadata stays in the JSON export.
Workspaces stay on your computer and are separated by Agentlas account. Sign out to switch accounts; your own graph is available when you sign in again on the same machine.
Separate publishing channels from source references
Left: choose a suggested publication or listing channel from its icon and short name. Hover for a brief summary and original link. The agent checks the channel’s public submission route, rules and feed before including it; if no suitable route is found, no channel recommendation appears.
Middle: explore the whole map, select a post to focus its source path, or select a channel to view the icon route: brand → channel → public feed → possible AI discovery → creators. Solid links and dashed candidate connections stay distinct. The suggested route remains an estimate.
Right: browse source names and compact references, with a short summary and original link. A project homepage, repository or paper is a reference original, not automatically a place to submit new product information.
- Discover related social content through several keyword angles; compare visible views within each platform.
- Trace intermediate and earliest-found sources while preserving unknown dates and counts.
- Evaluate actual publication/listing routes; selecting a suggestion does not submit, publish or contact anyone.
Keep recommendations within the evidence
Connections are research estimates, not proof that a creator read a source, used an AI collector or copied it. “Earliest found” is limited to this sample. The compact screen omits confidence and evidence explanations; the JSON export retains the underlying research metadata.
Every suggestion remains unexecuted. Submission rules, account requirements, self-promotion limits and temporary restrictions must be checked each time. A channel that forbids the proposed activity or is closed should be omitted or investigated with that restriction stated. The app does not assign exposure probabilities or promise acceptance, indexing, creator discovery or reach.
Missing views, publication dates and inaccessible platform results remain unknown or partial. AI-reported observations remain distinct from the app's own collection. No social post, listing or submission is completed by researching or selecting a recommendation.
The local workspace stores source metadata and selected links rather than copied article paragraphs. Your AI host and the public sites it visits still follow their own data policies; review research exports before sharing them.
Start your own source graph
Get the latest release or give the MIT-licensed GitHub repository URL to your local Codex or Claude Code chat. Ask it to install and open the app, sign in with Agentlas, and continue with your product URL. From the browser home, enter another URL whenever you want to start a new investigation.
