Capture Engine

What the Capture Engine Does

The Capture Engine is the entry point for all role knowledge in RolegacyAI. Its function is to ingest information from the environments where work actually happens — documents, meetings, email threads, project management tools, and workflow events — and convert that raw material into candidates for structured role memory.

Capture is not passive archiving. Every piece of content that enters the system passes through a processing pipeline that extracts the operationally relevant signal: what was decided, what was learned, what stopped working, what relationship was established, what workaround was discovered. Content that carries no institutional signal is discarded before it reaches storage.

Input Sources

The Capture Engine is designed to work across the full range of environments where role knowledge is generated:

• Documents and files: Reports, process guides, project retrospectives, vendor agreements, architecture decision records, and operational runbooks — any document that a role holder produces or consumes as part of executing their responsibilities.
• Meeting transcripts and recordings: Decisions made in meetings, escalations resolved in calls, and lessons surfaced during retrospectives are often never written down. The Capture Engine processes transcripts to extract these institutional moments before they disappear.
• Workflow and ticketing events: Signals from tools like Jira, ServiceNow, and project management platforms can trigger capture: project closures, escalation resolutions, milestone completions, and incident postmortems.
• Direct role holder input: Role holders can explicitly add memory entries — decisions they made, lessons they learned, configurations they discovered — directly through the RolegacyAI interface or via connected tools.
• Email and communication threads: Structured extraction from email and messaging threads where key decisions or agreements were reached and documented only in the exchange.

Trigger Mechanisms

Capture can be triggered in three ways: continuously (stream-based, always-on capture from connected tools), event-based (triggered by specific workflow events such as a project being closed or an incident being resolved), and manually (role holders explicitly initiate a capture session to document a specific decision or experience).

Trigger-based capture is particularly valuable during role transition periods, when the departure of an incumbent signals an urgent need to capture everything the outgoing holder knows before they leave.

The Processing Pipeline

Raw input passes through a multi-stage processing pipeline before any memory candidate is created. First, the content is parsed and chunked into semantically coherent segments. Each chunk is then assessed for institutional relevance — does it contain a decision, a lesson, a workaround, a significant relationship, or a process pattern? Chunks that pass this filter are passed to the Classification Engine for categorisation. Chunks that do not carry institutional signal are discarded.

This approach keeps the memory store focused on what matters operationally, rather than becoming an undifferentiated archive of everything a role holder ever touched.

Handling Sensitive and Personal Content

Not everything captured is appropriate for the institutional role memory. Personal opinions, private conversations, performance-related content, and information that identifies individuals rather than role patterns are flagged during capture and routed to the Sanitisation Engine before any storage decision is made. The Capture Engine is the first line of the boundary between personal and institutional knowledge — a boundary that is foundational to RolegacyAI's privacy model.