Tessera’s retrieval and agentic capabilities are powerful but passive. They respond to queries. The personal API layer makes Tessera active: connected to the data sources that represent my life in motion.
The API does not connect to the internet. It connects to local services: my email client, my calendar, my file system, and the exports from the various systems I use. The connections are pull-based. Tessera pulls data from sources on a schedule I control. Nothing pushes to Tessera, and Tessera pushes to nothing.
The Integration Points
Email is the highest-volume integration. Tessera pulls from the local mail store every fifteen minutes, processing new messages through the enrichment pipeline. Emails that contain decisions, commitments, or client information are enriched fully. Routine emails (newsletters, automated alerts, scheduling confirmations) are enriched minimally to reduce processing load.
Calendar integration provides the temporal scaffolding that Tessera uses for scheduling analysis, conflict detection, and workload assessment. Calendar events are enriched with participant information from the graph, relevant prior decisions for recurring meetings, and commitment deadlines that fall within the event’s timeframe.
File system monitoring watches designated directories for new or modified documents. Technical documentation, proposals, assessments, and project artifacts that I save to specific locations are automatically ingested. The monitoring is rule-based: only designated directories, only designated file types, never system files or temporary artifacts.
The Batch Import Interface
For historical data, I built a batch import interface that processes large archives: email exports from prior decades, document collections from past engagements, and ticket system exports from various PSA tools I have used over the years.
Batch imports are the most demanding operation. A single email archive from 2008 contained forty thousand messages. Processing them through the full enrichment pipeline took eighteen hours. The knowledge graph grew by sixty thousand nodes in a single import. The vector index required a full rebuild to incorporate the new embeddings efficiently.
The batch import taught me something important about personal knowledge systems: the value of historical data is not linear. The most recent five years of data account for about seventy percent of the retrieval value. The prior eighteen years account for thirty percent. But that thirty percent contains the foundational decisions and patterns that give Tessera its longitudinal depth. Without the historical data, Tessera would be a competent short-term assistant. With it, Tessera understands how I evolved.
What Stays Out
Some data sources are explicitly excluded. Social media. Web browsing history. Phone call recordings. The exclusions are privacy decisions: data that is too intimate, too noisy, or too legally complicated to include. The boundaries of what Tessera knows should be deliberate, not accidental.
I review the integration boundaries quarterly. As the system matures and my trust in its security model increases, I may expand the boundary. But expansion should be gradual and considered. There is no urgency to give Tessera everything. The current data sources provide more than enough context for the use cases that matter.