Back to the archiveW—03 / 11

Personal / Open SourceSelected work03 / 11

Selected work / Product Designer & Full-Stack Developer

TheVault

Product ArchitectureDesktop Application DevelopmentDeveloper ToolingAI Agent Infrastructure

Scroll to enter

01The project, in one breath

Role: Product Designer & Full-Stack Developer
Client: Personal / Open Source
Year: 2026

Source Code Windows Releases

The first challenge was turning a growing local memory store into useful agent context instead of another data dump. The recall engine combines project matching, typed records, promoted decisions, recency, keywords, tags, pruning, and explicit open loops to return compact packs…

Local-first AI memory system for coding agents, with an Electron desktop app, CLI, and MCP server for durable project context, handoffs, recall, task queues, and Graphify-powered code intelligence.

02 / Project film

Terminal output showing an AI agent recalling Vault context and reading a project handoff — Vault MCP recall and handoff workflow

Terminal session showing Vault MCP recall results, open-loop context, and structured project fields — Structured recall verification in the terminal

The Vault Recall screen with efficiency metrics, candidate reduction chart, and compact recall log — Recall efficiency dashboard

The Vault Graphify screen displaying a colorful interactive project graph and community legend — Graphify project graph

The Vault Settings screen showing bundled MCP runtime setup and connection status for supported agent clients — MCP client connection setup

Vault MCP recall and handoff workflowStructured recall verification in the terminalRecall efficiency dashboardGraphify project graphMCP client connection setupScroll / cut

03The build, reduced to what mattered

Three decisions shaped the final experience.

DURABLE CONTEXT

Structured Project Memory

Typed records capture decisions, plans, sessions, handoffs, summaries, artifacts, references, related files, priority, lifecycle state, tags, keywords, and explicit next steps per project.

RECALL ENGINE

Smart Recall & Open Loops

Candidate search, ranking, pruning, promoted decisions, and pressure-ranked open loops produce compact context packs while keeping unfinished work visible and resolvable.

MCP INTEGRATION

Cross-Agent MCP Continuity

The vault-memory MCP server gives Codex, Claude Desktop, Claude Code, and other MCP clients one shared interface for recall, saves, tasks, project hygiene, and handoffs.

OutcomeWhat remained after the build

The Vault delivers one local operational memory layer across desktop, CLI, and MCP clients. A coding agent can save a structured handoff in one session, another client can recall the same decisions and next steps later, and the operator can inspect or curate the record from the desktop console. The project also extends continuity beyond text memory: Graphify narrows repository exploration with structural context, while Vault Collab coordinates live sessions and handoffs without silently spawning workers. Local…The Vault delivers one local operational memory layer across desktop, CLI, and MCP clients. A coding agent can save a structured handoff in one session, another client can recall the same decisions and next steps later, and the operator can inspect or curate the record from the desktop console. The project also extends continuity beyond text memory: Graphify narrows repository exploration with structural context, while Vault Collab coordinates live sessions and handoffs without silently spawning workers. Local…

Electron desktop, CLI, and MCP serverInterfaces

Local-first and inspectableStorage Model

MIT open sourceLicense