The work is still there. Masthead makes it findable.
Masthead is a local-first session data layer for AI coding agents. It captures work from supported harnesses, keeps it in SQLite on your machine, and turns selected sessions into published knowledge artifacts—dossiers, runbooks, ADRs—that people and agents can search and reuse.
Agents create valuable context. Most teams throw it away.
Every agent session contains decisions, tool calls, failed paths, implementation details, and project history. That history usually stays trapped in one harness, one transcript, or one developer's memory.
Masthead treats that history as a product data layer. Sources import harness sessions, Workbench turns raw work into quality-checked published artifacts, Logbook holds only what you publish, and read-only MCP hands the next agent answers with sources—not a private log dump.
Sources, Workbench, Logbook, MCP, Now.
The boundary is the product. Masthead keeps capture and storage on machines you control, gives people a path from raw session to published knowledge, and exposes selected artifacts through read-only retrieval—not a monitoring tower or session library.
Sources
Discover and enable live harness connectors so sessions land in one local record.
Local SQLite store
Your machine holds the canonical record—sessions, sources, artifacts, and audits stay local.
Workbench and Logbook
Workbench: raw session → quality → agent authoring → publish. Logbook: published artifacts only, not a session library.
Read-only MCP
Other agents query published knowledge through bounded tools; they do not write your store.
Desktop runtime
Electron, TypeScript, and React—macOS and Linux builds via GitHub Releases (Windows on the roadmap).
Now
Shallow live presence across supported harnesses—observability as a view over collected data, not a tower.
Masthead is a developer tool and a systems case study.
For employers and clients, the signal is not only that Masthead exists. The signal is the judgment underneath it: recognize the hidden data layer inside AI work, design privacy boundaries first, and build a tool that improves future agent sessions without centralizing private work history.
Published knowledge and retrieval as the product surface.
Capture, store, and publish stay on machines you control.
Electron app, SQLite, Workbench, Logbook, MCP, and site.
Sources → Workbench → Logbook → MCP—not a vague dashboard.
Need AI work to keep the record?
Masthead is the public proof of how I think about agentic systems: local-first data, clear boundaries, and published knowledge over demo theater.
Email Tyler