Knowledge Base

The problem

The answer exists. Someone wrote it in a ticket reply eighteen months ago, or it is in a Word document on a share drive, or it is in a technician’s head. So the same question arrives again, and someone answers it again, and the answer is just as unfindable afterwards as it was before.

For an MSP the filing problem comes with a disclosure problem. The runbook for a client’s firewall failover contains that client’s topology. The internal triage notes are not for clients at all. A single “publish” button forces a choice between writing nothing down and writing it somewhere the wrong reader can reach it — so most of it gets written down nowhere.

And when the provider does publish to a client, the client is handed something they cannot shape: an article filed under someone else’s taxonomy, appearing in their help centre whether it fits their business or not.

What Solidlio does about it

Every article carries a visibility level as well as a status: MSP-internal, one organization, one account, named client organizations, all managed clients, or platform-wide. That level is evaluated on every customer-facing read — Help Center, search, related articles, the AI assistant and ticket deflection all apply the same rule set from the same place, so a new reading surface cannot forget it. Receiving clients get the other half of the control: hold new shares for review, hide any article, or re-file it under their own categories, without touching the provider’s copy. Every content change is versioned and restorable. And published articles are embedded as vectors, so customers find answers by describing the problem — including in the deflection panel that runs while they are typing a ticket, before it becomes one.

Capabilities

Built for MSPs and their clients

A managed client is a separate tenant, not a folder. That is what makes one article publishable to three different audiences without copying it — and what lets the receiving client own its own presentation.

Three structural guarantees fall out of this. An MSP-internal article has no visibility value that a customer read path will match, so it cannot be surfaced by any customer-facing feature, present or future. A tenant-owned article cannot be given cross-tenant reach: an author who asks for platform-wide publication is scoped to their own account unless they are a platform administrator. And a client’s decision to hide or hold an article never edits the provider’s copy — the two sides cannot overwrite each other.

How it works

1. Write it — a real Markdown editor, in the portal you already work in.
2. File it — one of your organization’s categories, a type and tags.
3. Choose who reads it — internal, one organization, an account, named clients, all clients, or platform-wide.
4. Publish it — the article is embedded for semantic search on the way in, and who published it and when is recorded.
5. The client decides how it lands — published straight away, or held for their review, hidden, or re-filed under their categories.
6. It finds the reader — Help Center, search, the AI assistant, and the deflection panel on the new-request form.
7. Readers tell you if it worked — helpful or not, with an optional comment.
8. Edit freely — every change is versioned, and any earlier version reads in full and restores without losing what replaced it.

Compliance and audit

• Every content change is retained. The superseded text, who replaced it, when, and an optional reason — so “what did this policy say in March” is a question with an answer.
• Publication is attributable. The publisher and publication time are recorded on the transition, not only when an article is created published.
• Every feedback submission is written to the audit log with the article title, the verdict, the comment and the submitting person — alongside the queryable row that drives analytics.
• Cross-tenant reads are indistinguishable from missing articles. A request for another tenant’s article returns “not found”, never “forbidden”, so identifiers cannot be enumerated by trial.
• The tenant is taken from the authenticated session, never the request body. Search and deflection accept an organization field for backward compatibility and ignore its value.
• The role matrix is asserted, not assumed. A test walks all six roles against the authoring routes, so a change to the access floor fails the build rather than shipping.

Editions

The knowledge base carries no plan gate. Every capability above is available on every tier, including Free.

Indexing an article for semantic search costs one AI credit, drawn from the account’s existing allowance, so plan size governs how much of a corpus is searchable at once rather than whether the feature exists. Writing, versioning, publishing and browsing are never refused for want of credits — an article that cannot be indexed saves anyway, says so, and can be re-indexed later. Semantic queries are not separately charged; Sol conversations are metered per turn.