Theka v1.2.0; Knowledge Layer and AI Wiki: your documents' knowledge, organised

Knowledge Layer: your documents' knowledge network

• Theka automatically analyses every uploaded document and extracts recurring concepts (people, companies, products, regulations, projects). Each concept becomes a Topic with its own dedicated page.
• Every Topic collects the verifiable facts extracted from the documents that mention it. Each fact is linked to its source document and to the AI's confidence level.
• Topics are connected to each other in a navigable knowledge network: contract X is signed by person Y who works for company Z. Explore the network starting from any Topic and follow relations across multiple hops.
• Promote to entity: turn an AI-extracted Topic into an officially curated entity with one click, keeping all the accumulated knowledge.
• Merge Topics: if the AI created two Topics for the same concept, merge them into one with full history preserved.
• Import from existing entities: a bulk action creates Topics for the entities you already manage, without starting from scratch.

AI Wiki for every concept

• Every Topic carries an auto-generated wiki page that synthesises everything Theka knows about that concept, reading facts extracted from dozens or hundreds of different documents.
• The wiki updates automatically when new documents mentioning the Topic arrive, flagging itself as needs regeneration right away.
• Every wiki statement carries its source documents: users can always trace back to the originals.
• Visibility-aware: users only see facts extracted from documents they can access. The wiki tailors its content to the reader's permissions.

Knowledge tab on the document detail

• New panel on every document detail showing the extracted Topics, the number of facts collected and the AI analysis progress in real time.
• Manual re-run of knowledge extraction on a document — handy after edits or re-uploads.

Smarter related documents

• The Related documents panel in the asset detail now also factors in shared concepts and knowledge graph relations: it finds linked documents even when they don't share entities or tags, but talk about related concepts.
• The relevance score is richer, summarising more proximity signals between documents in a single view.

Files linked to documents

• Attach additional files of configurable type to any document version: signed copy, preview, translation, free attachment.
• Attachment types are tenant-customisable: the admin decides which types to make available and which should skip AI analysis.
• New aggregated attachments view that shows every file and related document for an asset at a glance.

Bulk actions on selected documents

• From the document grid you can select multiple assets and assign them to a project or link them to an entity in one go.
• The action is additive: existing projects and entities stay attached, new ones are appended.

Concept visibility controls

• Three admin-configurable visibility modes: standard (Topics reflect document permissions), transparent (users see that hidden facts exist without seeing their content), strict (Topics with no accessible source are fully invisible).

• Fully redesigned landing page: new visual identity, rewritten copy explaining more clearly what Theka does and who it's for, supported-integrations section that updates in real time and a quicker demo-request flow.
• AI generation status visible on the document detail: a new label shows whether knowledge analysis is running, done, queued for retry or skipped — and why. No more guessing why a document has no AI data yet.
• Knowledge layer AI cost controls: the admin can set a daily token cap and a priority (normal, low, paused) on knowledge analysis, independent from other AI operations.
• Per-category knowledge toggle: concept extraction and wiki generation can be enabled or disabled per document category, without turning the feature off globally.
• The tenant members list is now visible to all users, but with essential fields only (name, email, role): collaboration is easier while admin details stay reserved for admins.
• Public supported integrations catalogue: new marketing page that shows in real time every storage, AI, search engine, ERP and SSO connector with vendor logos and descriptions.
• Auto-create Topic from entities (opt-in): when on, every new entity automatically spawns a linked wiki Topic.
• Knowledge backfill on the historical archive runs on a dedicated queue, so it doesn't slow down upload and analysis of today's documents.
• On a Topic page, dedicated buttons to generate the wiki at first access and regenerate it after changes, with clear progress feedback.
• Sturdier approval and signing workflow: the approval-stage configuration page now validates every combination in real time, preventing inconsistent flows that would leave documents stuck. Deleting a step checks its impact on other stages first.
• Anomaly banner on the document detail: if the document type's approval workflow has inconsistencies (e.g. leftover stages from a deprecated setup), an amber warning surfaces it to the admin at a glance, without having to dig into the configuration panel.
• Signed document in the attachments section: the file uploaded as the signature of an approval now also appears in the unified Attachments section of the document, alongside other attachment types, sorted by category.

• Fixed a case where Topic wiki regeneration could restart endlessly if the AI configuration for the knowledge layer was incomplete: users now see a clear message about what to configure.
• Stabilised database connection behaviour during low-load periods: removed rare temporary stalls that could appear after long idle stretches.
• Refined related document relevance: scores are now consistent between the asset panel and the administrative graph view.
• Improved AI analysis robustness when a tenant exhausts its daily token quota: the document is queued for the next day instead of silently failing.
• Knowledge extraction no longer re-fires redundantly when the module is off for the tenant or for the document's category.
• Fixed a case where a document could move into the Signed status without the signed file actually being uploaded: the document now only advances after the signed upload.
• Fixed a case where a bulk project or entity assignment showed a success confirmation but did not actually link anything: the link is now persisted, or the user sees a clear error.
• Improved the error message when trying to skip an approval step: it now explicitly states which step must be completed before advancing.
• Fixed a case where an approval workflow configuration could enter a state with no recoverable path through the UI, forcing manual intervention to unblock affected documents.

Major release: introduces the Knowledge Layer, the biggest step forward in Theka's history. Your documents are no longer just searchable files: they become a navigable knowledge network with concepts, facts, relations and AI-generated wiki summaries.

The module is included in Professional and Enterprise plans: the admin finds it on the tenant configuration page. Customers staying on the base plan keep using Theka exactly as before — no existing feature changes or requires action.

No action required on already-uploaded documents: as soon as the module is enabled, knowledge analysis of the historical archive starts in the background with low priority, without impacting daily activity. The admin can adjust the daily cost cap and pause analysis at any time.

All pre-existing features (upload, search, approvals, sharing, integrations, audit) remain 100% unchanged.