A local-first memory and operational awareness system for Claude Code workflows and Windows focus tracking.
Overview
Locus combines two ideas:
- persistent project memory
- operational focus tracking
The system captures and resurfaces high-level project context across Claude Code sessions while also tracking real operational activity on Windows over time.
Instead of storing full transcripts, Locus records structured concepts such as:
- architectural decisions
- implementation summaries
- project conventions
- rejected approaches
- workflow notes
- operational context
Alongside that, the runtime continuously tracks focus activity including:
- active applications
- focus duration
- session timing
- operational patterns
- historical activity trends
The goal is simple:
Important project context and operational state should not disappear between sessions.
Claude Code integration
Locus integrates directly with Claude Code through MCP.
This allows Claude workflows to:
- search project memory
- review previous decisions
- inspect earlier implementation context
- recover workflow history
- continue long-running work more easily
Locus does not modify the model.
It provides a persistent local context layer that developers can review and reuse across sessions.
Focus tracking
Locus also operates as a persistent Windows focus tracking runtime.
The system records operational activity across:
- sessions
- days
- weeks
- longer-running workflows
This creates a lightweight historical operational surface showing:
- what received attention
- how long focus lasted
- what applications were active
- how work patterns changed over time
The focus system is intentionally passive and local-first.
It is designed to expose operational reality rather than create productivity theatre.
Memory model
Locus focuses on structured contextual recall rather than transcript persistence.
Captured memory remains:
- local-first
- inspectable
- reviewable
- searchable
- developer-controlled
The emphasis is on preserving useful project concepts and operational continuity rather than simulating persistent agent memory.
Diagnostics
Locus is designed to be diagnosable rather than opaque.
The system exposes tooling for:
- memory status
- project scanning
- import visibility
- recall inspection
- focus history inspection
- diagnostic workflows
Developers can see what has been captured, what is available and how the system is behaving over time.
Locus at a Glance
Core ideas
- Persistent local project memory
- Claude Code integration
- MCP-based tooling
- Structured concept capture
- Operational focus tracking
- Historical focus visibility
- Developer-reviewable memory
- Diagnosable system state
Technical stack
- Go runtime
- MCP server
- CLI tooling
- SQLite persistence
- Windows focus tracking
- Project scanning
- Local-first architecture
Why it is different
Most AI coding workflows lose context between sessions and most productivity tools fail to capture real operational activity.
Locus combines both:
- persistent project memory
- long-term operational focus tracking
The result is a lightweight local system that preserves both project understanding and historical execution state over time.
The memory belongs to the developer and the project rather than the model itself.
Relationship to earlier projects
Locus originally evolved from earlier operational tooling projects including CommandDeck and focus-reader.
The current version combines those operational awareness ideas with persistent memory tooling for Claude Code workflows.
Closing note
Locus is built around a simple idea:
important project context and operational history should remain accessible across long-running development workflows.
The focus is not transcript persistence or artificial agent memory.
The focus is preserving useful project-level concepts and operational visibility in a structured local system that developers can inspect, search and reuse over time.