Loading Your Company Context | TurnkeyGTM Support

Every agent in Upstream works better when it knows who you are. The Company page (/company) is the permanent home for your company profile, brand and context. Guided setup and review use /company/onboarding, with the same saved information and stages. Old /onboarding links redirect there, preserving the selected stage. Both addresses read the same saved information; no copy or migration is needed. Save your changes before switching between these addresses — unsaved drafts stay with the page you are editing.

What discovery does

Enter your website (a company name is optional — we find it). Most runs finish in well under a minute; a large site can take up to two. Upstream:

Reads your homepage — title, description, navigation, structured data, logo and social links.
Finds your sitemap — via robots.txt, then the usual sitemap locations.
Classifies key pages — blog, docs or help center, pricing, about, features, solutions, integrations, customers, security, changelog and more. A page's own address counts for more than a matching subdomain name, and if one of them turns out to redirect somewhere that is not what it claimed, we use the next candidate or leave that slot empty rather than guess.
Reads a handful of pages — about, features, solutions, pricing, customers — never more than fourteen, and only from the site your address resolves to.
Reads your CSS — brand colors, fonts and corner style, with the place each one was found.
Drafts your profile — one AI pass over what was read, with a confidence score.

The live feed on the right shows each find as it happens, and under each one what an agent can now do with it (for example, a blog URL is enough for the content marketer to start exploring).

Discovery reads pages only from the site your address resolves to and its subdomains — if what you entered redirects anywhere else — including up from a subdomain to your main domain — the feed names the site we read instead (plus the stylesheets those pages link, which often sit on a CDN), follows at most a few redirects and checks every one of them against internal networks, uses only standard web ports, respects your robots.txt, and counts as one AI call on your plan. One run can be active per workspace at a time, up to six runs an hour.

Confirming your profile

The Company, Offering & key pages and Brand stages show everything discovery filled in. Next to each value is a chip that says where it came from:

You — you typed it. Discovery never overwrites these, even if you run it again.
Discovered — read from a page or your CSS, with a confidence score.
AI draft — proposed by the AI pass over your pages.

Empty fields marked Recommended explain why a downstream agent wants them and are ordered by the role you chose in the welcome wizard. You do not need to fill everything: a name, a one-line description, your target customers and your products already make a big difference.

Products, features, use cases and competitors are lists you can edit. Untick Include on any row to keep it on your profile but out of the graph.

Switching stages keeps your unsaved Company, Offering, Brand and Memory drafts while this page remains open in the same workspace. Save profile changes or import memory before reloading, leaving the page or switching workspaces. If you choose another stage while Save and continue is pending, its completion will not take you away from your newer choice.

Adding company knowledge

The Company knowledge stage previews the company details that will be added: your organization, products, capabilities, use cases, competitors, messaging and target customers. Choose Add to company knowledge to save them in Knowledge with an Onboarding badge. After changing your profile, choose Update company knowledge. This updates items that are still unchanged since they were added. If you edit an item in Knowledge, its content, metadata and relationships are left alone, with a warning; the same protection applies to older items whose ownership cannot be verified.

Optionally, select Include discovered pages as sources to make the pages we read searchable, so chat answers can cite your site. New information extracted from those pages may appear in Review.

If you remove a row (or untick Include) and update company knowledge again, the matching unchanged item is removed only when a workspace owner runs the update. A member's update reports the rows an owner still needs to clear. Items edited in Knowledge are not removed, even for an owner.

Only one company-knowledge update can run per workspace at a time. If one is already running, wait for it to finish before retrying. An update can partially complete when storage or search indexing is unavailable. Running it again retries indexing of the latest saved revision of items tracked by this profile, including items you edited in Knowledge, without overwriting those edits. Failed removal/indexing work remains tracked for retry. Approved items remain available through keyword and graph search while vector indexing needs repair.

Bringing in your AI memory

If you have been using ChatGPT, Claude, Gemini or Copilot for months, they already know a lot about your company. The Memory stage gives you a prompt to paste into that assistant; paste its answer back and it is imported like any other document. Each import is listed with a link to its source. You can prepare another draft while an import is pending; finishing the earlier import will not clear newer edits to the text, title or assistant selection.

Connecting sources

The Sources stage lists where company knowledge usually lives. GitHub Issues connects today (an owner-managed, read-only token). Everything else is marked Planned with the path that works now — export a document and drop it on the Upload page. Click I want this to tell us which connectors matter to you.

Sources imported with review isolation keep new extraction in human review and keep their source text out of semantic chat context, including after reimport or reindex. A later import cannot silently remove that restriction. Requesting isolation for an existing source reprocesses it even if its text is unchanged; wait for any current job to finish first. Previously extracted Knowledge entities are not automatically removed.

Ready

The Ready stage shows your context readiness, the fields still worth a minute, and what you can do next for your role. Ask a first question right there — answers can cite the company knowledge you added.

Troubleshooting

"Only standard web addresses" or "internal networks" — discovery works on public http/https sites on ports 80 and 443. Internal hosts, private-network addresses and custom ports are refused, including when a redirect points at one.
"A discovery run is already in progress" — one run per workspace at a time; wait for the feed to finish (under two minutes, whatever the site) and try again.
A field will not save — web addresses must start with http:// or https://. A bare domain such as acme.com is accepted and completed for you; anything else is rejected with a message naming the field.
Discovery ran out of time — very large sites hit the fetch budget. Run it again, or add the key pages by hand; everything already saved stays.
A seed or source stays in progress after an interruption — do not repeatedly restart it. Contact support with the workspace, source/job and, when shown, Support attempt identifiers; do not include pasted source text. Support must stop new work, verify every writer and outstanding storage/indexing operation has finished, and review partial results before recovering the exact abandoned attempt. There is no automatic age-based restart.
Source changed or reindex refused — another request may have reserved that source. Wait for its current job to finish, then retry. An unchanged import can resume a queued job after a lost response; a different version cannot replace an active writer. Reindex errors show the server's recovery guidance. If the job remains interrupted, use the support process above.
Stored source/job review policies disagree — processing is stopped rather than using the weaker policy. Contact support with the source/job identifiers; do not remove review flags or keep retrying to bypass the conflict.