SourceHarbor

Project Positioning

SourceHarbor is a source-first engineering repository for AI knowledge intake and MCP-ready long-form source operations.

What that means in plain language:

• the repository is meant to be cloned, inspected, and run
• the strongest proof is local runnable evidence
• the project is not marketed as a turnkey hosted SaaS

The product shape is still real:

• API for ingestion and inspection
• Search / Ask front doors over grounded retrieval
• Watchlists / trends / bundles as the compounder layer
• Worker for pipeline execution
• MCP for agent-facing tooling
• Web command center for operators
• Sample playground and use-case pages as truthful discoverability surfaces

The public promise should stay smaller than the internal engineering ambition. That keeps the repo honest.

Current Non-promises

SourceHarbor should still not be described as:

• a turnkey hosted team workspace
• a managed SaaS product
• an autonomous agent autopilot

Those directions now have explicit internal ledgers instead of implied roadmap copy. Public docs keep only the stable summary here; the full working contracts stay in the internal planning ledger used by maintainers.

Two directions are intentionally kept in the “evaluate before promise” bucket:

• Agent Autopilot: worth a human-in-the-loop spike, not ready for product-level claims yet
• Hosted workspace: worth a readiness study, but still incompatible with the current source-first and local-proof-first promise if overclaimed

The same principle also governs builder packaging:

• the repo-local CLI discoverability surface is real now
• thin public CLI / SDK surfaces are now real
• public starter packs are now real, but still first-cut
• plugin-first positioning stays no-go now

If you want the durable ship-now / later / spike-only / no-go ledger for CLI, SDK, builder ecosystem, hosted, autopilot, and plugin-style expansion, read ecosystem-and-big-bet-decisions.md.

Future-Direction Boundaries

These two directions remain real topics, but they are not current public product promises:

• Agent Autopilot: worth a human-in-the-loop exploration slice; not ready for shipped-capability copy.
• Hosted workspace: worth a readiness study; still incompatible with the current source-first and local-proof-first promise if overstated.

Public docs intentionally stop here. If a maintainer needs the full working contract, the canonical internal copy lives in the internal planning ledger. Keep those blueprints internal; the public narrative should always stay summarized in this document and the other public ledger docs, not lifted directly from the internal execution contracts.

This site is open source. Improve this page.