Status: design spike only. This document is a readiness assessment, not proof that SourceHarbor is already a hosted or managed product.
Think of this document like a building inspection, not a sales brochure. SourceHarbor already has real rooms, working plumbing, and a front door. It does not yet have the hotel-grade locks, operations desk, and service contracts a hosted team workspace would need.
Short answer:
That current position is a product strength. Turning it into a hosted story too early would make the promise larger than the proof.
| Question | Verdict | Why |
|---|---|---|
| Can SourceHarbor honestly be described today as a hosted SaaS? | No | the public proof model still centers local runtime truth and explicit secret ownership |
| Is there enough product shape to study a managed slice later? | Yes | search, ask, MCP, ops, bundles, and playground already look like a coherent product surface |
| Should the first hosted-shaped experiment be a shared team workspace? | No | multi-tenant auth, isolation, custody, and remote support are not ready |
| Is there a smaller hosted-shaped slice worth exploring? | Yes, conditionally | a read-only or single-owner evaluation layer could improve activation without breaking source-first honesty |
These are the surfaces that already feel product-like today and what they mean for a future managed slice.
| Surface | Current repo truth | Evidence anchors | Managed-readiness signal |
|---|---|---|---|
| Front door | Search, Ask, MCP, Ops, and compounder pages already describe a coherent result path | README.md, docs/start-here.md, apps/web/app/search, apps/web/app/ask, apps/web/app/mcp, apps/web/app/ops |
good product shape |
| Operator workflow | command center, jobs, feed, watchlists, trends, settings, and bundles are visible product rooms instead of hidden scripts | apps/web/app/page.tsx, apps/web/app/jobs, apps/web/app/watchlists, apps/web/app/trends, artifacts/releases/README.md |
good operator-facing foundation |
| Agent workflow | MCP server exists on top of the same system rather than a fake parallel stack | apps/mcp/server.py, docs/mcp-quickstart.md |
good agent-facing foundation |
| Local runtime orchestration | repo-managed boot path, doctor path, compose stack, and systemd examples already exist | docs/start-here.md, infra/compose/core-services.compose.yml, infra/systemd/sourceharbor-api.service |
useful ops building blocks, but still self-host shaped |
| Proof and trust surfaces | proof ladder, runtime truth map, and public positioning already reject overclaim | docs/proof.md, docs/reference/project-positioning.md, config/public/github-profile.json |
strong honesty layer |
| No-boot evaluation | playground, sample corpus, and use-case pages already support product evaluation without full setup | docs/samples/README.md, apps/web/app/playground, apps/web/app/use-cases |
good seed for a managed evaluation slice |
This is the real gap table. In plain language: these are the parts where “someone else runs it for me” becomes much harder than “the code can run.”
| Area | Current truth | Why it blocks a hosted workspace promise |
|---|---|---|
| Identity and isolation | there is no tenant model, tenant-aware auth contract, or workspace partitioning surface | a hosted team workspace needs strong separation between users, data, and actions |
| Secret custody | notifications, Gemini-backed review, and some live-provider flows depend on operator-held secrets | a managed offering would need clear rules for secret storage, rotation, exposure, and abuse containment |
| Runtime proof boundary | public proof explicitly distinguishes local proof from remote proof | the current repo is honest because it does not pretend remote reliability is already proven |
| Onboarding model | docs still start from clone, .env, uv sync, Docker/compose, and local route snapshots |
that is repo-native onboarding, not hosted-native onboarding |
| Operational durability | API, worker, Temporal, and Postgres are real, but not documented as an SLO-backed managed service | hosted users would expect uptime, restart behavior, retention, and support boundaries |
| Artifact and data policy | evidence bundles and job artifacts exist, but hosted retention, privacy, and storage lifecycle are not a documented contract | managed storage is not just “save files somewhere”; it needs policy |
| External-provider dependence | YouTube, notifications, and Gemini-backed lanes still rely on provider and secret availability | a hosted product would inherit provider outages and quota pressure as product responsibility |
SourceHarbor is strongest today as:
That matters because the current trust story is: “you can verify the machine yourself.” A hosted story changes the sentence to: “trust our managed operations.” That is a much bigger contract.
If SourceHarbor later experiments with hosted delivery, the first slice should be:
managed evaluation and observability for a single-owner environment, not a shared hosted workspace
Recommended first slice:
| Included | Why it is small enough to stay honest |
|---|---|
| hosted front door and docs handoff | helps discovery without claiming full hosted operation |
| hosted sample playground | gives faster product understanding without taking arbitrary user inputs |
| hosted proof / docs / MCP quickstart handoff | makes the learn-before-install path smoother |
| read-only job trace / log viewer for demo or single-owner runs | shows operational visibility without crossing into multi-tenant execution |
| managed bootstrap diagnostics | helps activation without promising full platform operations |
Explicitly out of scope for the first slice:
Do not reopen full hosted workspace work until these are true:
These sources matter because they describe the difference between “an app can run” and “a managed system can be trusted.”
SourceHarbor implication:
| Direction | Recommendation | Meaning |
|---|---|---|
| Full hosted team workspace | No-go now | do not promise shared managed operation |
| Thin managed evaluation slice | Conditional go later | only as a single-owner or read-only bridge into the real local product |
| Hosted-flavored landing or docs surfaces | Go with honesty | safe only if they keep pointing back to source-first and proof-first truth |
Stop the hosted track immediately if any of these happen:
Do not say:
Do say:
If someone reopens this topic later, they should answer these questions first:
Until those answers exist, keep the public promise where it is: