FAQ

Both, honestly. Enterprise and mid-market teams usually come through us because they want a rollout conversation and their own naming. Startups and smaller teams can self-serve a workspace today under our design partner program. Start wherever fits; you can always reach out.

Public plans are on Pricing: Free ($0), Builder ($49/mo), Team ($149/mo), and Enterprise (contact sales). Workspaces have a credit balance and ledger for chat, MCP, and source index work; credit pack top-offs use Stripe Checkout. Design partner access may still be free at our discretion during early access (see our Terms of Service). SSO, SCIM, and compliance packaging are Enterprise and sold through sales.

From the dashboard, invite each person by email and choose their role. The invitation carries an accept link that shows them which workspace and role they are joining, signs them in, and lands them in the workspace. You can also copy the accept link and send it yourself, and revoke any pending invitation. See How to invite teammates.

The sign-in and account-creation pages stay available if we cannot confirm an existing session. Refresh after the service recovers and a valid session will send you to the dashboard. If a new sign-in still fails, retry your magic link or contact us.

Two different jobs. Go live: create a push token under Publishing credentials, then CI POSTs a package (a successful push becomes live), or promote a package from Sites. Ground answers: connect Sources and Sync now so chat can use the knowledge index; that does not flip the live site by itself. See How to publish from your pipeline, How to connect a source, and the quick start.

Push tokens for CI only. They start with hdpt_ and are shown once. They authorize package uploads that make the public site go live. They are not visitor chat credentials and are not used by Sources sync.

Yes. From the dashboard Sources panel you can connect GitHub (including private repos with OAuth), a documentation URL, Confluence, Google Drive, Slack, Jira, DocsGen, and OpenAPI. Sync indexes knowledge for grounded chat and MCP. Status is honest (Indexed, Never indexed, Busy, Sync failed, Paused). Soft skips such as already running show as Busy, not a silent no-op. ACL-bearing sources keep private material off the public widget. Some connectors need platform OAuth credentials enabled by us before Connect is available. See How to connect a source.

The dashboard shows honest status. Manual Sync now bypasses silent rate limits and can re-index even when the remote content hash looks unchanged. If another run is already in progress, you see Busy with a clear message, not a silent no-op. Transient start failures are retried automatically, and stale locks recover without dropping the sync request. Hard connector errors stay red with a short fix hint. Soft skips are not counted as needs-attention failures on Overview health.

No. Sources ground answers only. Sync updates the knowledge index for chat, MCP, knowledge search, and the docs graph. Going live is a CI push with a publishing credential, or an explicit Promote from Sites. That separation is live product behavior today.

Yes. In Sites you can manage multiple sites per workspace (plan max applies), set a primary host, promote a package version live, point a *-preview host at a package without cutover, and run sticky A/B against a partner site package. Custom domains remain coming soon.

Yes. DocsGen turns a GitHub codebase into draft documentation pages grounded in source files. OpenAPI generation takes an API description and produces a validated OpenAPI 3.1 document plus a deterministic TypeScript starter client. See Generation.

No. Keep authoring and building docs with the tools you already use. We only receive the output of your build, so your writing workflow does not change.

Each publish is a new immutable version, and the previous version stays reachable in storage until the next publish. A bad build is a re-publish away from fixed: push a known good build again and it becomes live the moment it validates.

It answers only from your published docs and cites the page behind every claim. Before an answer is shown, its claims are checked against your documentation, and anything that cannot be verified is left out rather than presented as fact. We design for verifiability rather than making absolute promises. See About grounded answers.

It says so instead of guessing, and the question is logged as a content gap. In Content gaps you can search the knowledge index, copy the question, open Sources, or open Sites when you are ready to promote. Export CSV is available. See Analytics and About grounded answers.

We search your knowledge index before calling a model. Strong matches can answer with zero AI tokens. Harder questions use AI with citations and use workspace credits. Chat quality and AI usage show the 30-day mix. See About grounded answers.

Retrieval matches questions by exact wording and by meaning together, every citation is checked against the source text before an answer is shown, and an automated evaluation runs nightly against the live product and fails loudly if grounding or citation accuracy drops. Graph-aware expansion can follow related pages for multi-hop questions. See About grounded answers.

Yes. The public site chat only retrieves content marked public. Signed-in workspace members can use private docs chat from the dashboard, which filters to public content plus any restricted ACL keys their linked source identities may see. See About grounded answers.

Yes. Each tenant site exposes a public MCP endpoint at /api/mcp with tools to search and fetch pages. Tenant isolation is hostname-derived, the same as the widget. See How to expose docs over MCP.

No. On hosted.devdocs.ai, the chat bubble answers only from our product FAQ database for visitors who are not signed in. It does not call a language model. Workspace subdomains use the grounded AI assistant after you publish.

Yes. An owner or admin can style each site manually or review colors, a curated font, and a logo found by a bounded scan of a connected GitHub repository at a pinned commit. The scan never runs repository code. Nothing applies until you choose the candidates and confirm them. See How to brand a site.

Not in the current design partner program. Every workspace is served at your-subdomain.hosted.devdocs.ai. Custom domains are a future consideration, not a current feature. See About workspace domains.

Everything a security or procurement reviewer usually asks for is public. Our security page states plainly what is available now, in progress, and on the roadmap, alongside our no-training commitment. The data processing agreement and the subprocessor list (Cloudflare, Anthropic, Better Stack, Sentry, Stripe) are published next to it, and your own workspace has a tamper-evident activity log you can export for a reviewer.

No. We do not train, fine-tune, or otherwise improve any model on your documentation or your visitors' questions, and our agreement with our model provider does not permit them to either. See Security.

Not yet. Single sign-on is on our roadmap. It is not available today, and we will not commit to a date until we are confident we can hit it.

Not yet. We are pursuing SOC 2 Type II, with completion targeted for late 2026. This work is in progress: we are not yet certified or audited. A tamper-evident activity log is part of that foundation.

Not during the design partner period. We work to keep the service reliable and will say plainly if that changes before it affects you.

The roadmap uses three labels. Live means you can open it and click through today, including the dashboard, sources, sites, reviewed repository branding, publishing, insights, credits, public docs, and trust center. Coming soon is in progress but not fully clickable yet, including custom domains and enterprise SSO with SCIM. Planned is future direction, including additional connectors and SOC 2 Type II targeted late 2026. We do not mark internal test suites as Live.