Reduce friction. Preserve context. Stay in control.
Friction Lab is an independent software studio building local-first tools for modern software engineering.
Our work focuses on the layers around development that are often fragmented across terminals, editors, AI agents, documentation, and the operating system itself. We build software that brings structure to those workflows without taking control away from the developer.
Across our products, the goal is consistent: make complex engineering work easier to navigate, reason about, and execute through deliberate architecture, deterministic systems, and practical automation.
Friction Lab products are built around a shared set of engineering principles:
- Local-first — source code, project context, working state, and engineering data stay on the user's machine wherever possible.
- Human-centered AI — AI can reason, plan, and execute, but consequential decisions remain visible, bounded, and ultimately controlled by the developer.
- Deterministic by design — workflow authority, safety boundaries, and critical state transitions should be explicit and reproducible rather than delegated to opaque model behavior.
- Architecture over shortcuts — systems are designed to remain understandable, extensible, and maintainable as they grow.
- Safe by default — interactions with source code, repositories, developer environments, and local files are deliberately constrained and recoverable.
- Purpose-built — every product begins with a concrete source of friction and earns its complexity by solving it well.
Local-first agentic software development platform
Relay is a structured execution environment for AI-assisted software development.
It coordinates agents across the full lifecycle of a development mission: translating human intent into actionable work, supervising execution, capturing evidence, preparing independent review, recovering from failure, and determining when work is genuinely complete.
Relay deliberately separates agent intelligence from workflow authority. Agents are free to reason and implement within their assigned roles, while Relay retains deterministic control over state transitions, automation permissions, evidence requirements, review readiness, and completion criteria.
The result is an agentic development system designed not around unrestricted autonomy, but around accountable autonomy.
Highlights:
- Mission and task orchestration
- Supervised autonomous agent execution
- Deterministic workflow control plane
- Scoped automation and approval policies
- Git preflight and automated evidence capture
- Independent reviewer preparation and handoff
- Guarded workflow actions and recovery paths
- Decision checkpoints and completion assessment
- End-to-end workflow integration testing
- Local-first persistence and native macOS integration
- Experimental local-model reasoning and evaluation
- Relay-on-Relay self-development infrastructure
Local-first knowledge infrastructure for software projects
Code Atlas gives software projects a durable layer of structured knowledge that lives alongside the codebase.
It connects documentation, engineering decisions, architectural context, source references, and long-lived project memory within a Markdown-native workspace. Rather than allowing important reasoning to disappear across conversations, issue trackers, and temporary notes, Code Atlas preserves the context required to understand how a system evolved and why particular decisions were made.
Its role is not simply to document a codebase, but to make project knowledge navigable, reusable, and durable.
Highlights:
- Project-aware Markdown workspace
- Structured project and source references
- Durable local project memory
- Code-aware navigation
- Reusable SwiftUI architecture
A local-first navigation layer for your working environment
Wayfinder is built around a simple idea: finding something should not require remembering exactly where you left it.
The project began as a fast Rust command-line navigator for moving through deeply structured filesystems. That CLI remains the foundation available today, combining interactive navigation, filesystem traversal, history, and shell integration.
Wayfinder is now expanding beyond the terminal into a broader contextual navigation system.
Wayfinder for Alfred and unified contextual search are currently in development, with the longer-term goal of making files, folders, projects, application windows, browser tabs, browser history, bookmarks, and saved working contexts discoverable from a common interface.
Instead of requiring users to decide where to search first, Wayfinder is designed around the destination itself.
Search for a project and the relevant result may be its folder on disk, an existing Finder window, an open development environment, a browser tab, or a previously saved workspace. Wayfinder's responsibility is to surface the useful destination and help the user get there.
Highlights:
- Interactive Rust command-line navigation
- Fast file and folder traversal
- Recent project history
- Shell integration
- Wayfinder for Alfred in development
- Unified contextual search in development
- Planned application and window discovery
- Planned browser tab, history, and bookmark search
- Planned saved workspaces and context restoration
- Local-first architecture
- Release packaging and automated testing
Friction Lab is focused on developer tooling, AI-assisted software engineering, contextual navigation, and durable project knowledge.
For work in digital content authenticity, invisible watermarking, and cross-platform media verification, see Roadscript, whose flagship application is Cyphra.
Software development is full of small interruptions that rarely look important in isolation.
Finding the right project. Recovering a working context. Reconstructing why an architectural decision was made. Verifying what an AI agent actually changed. Repeating setup that should have been remembered. Moving information between tools that understand only one part of the work.
Those costs accumulate.
Friction Lab builds software for those gaps.
We believe capable tools should reduce the amount of machinery developers have to think about while increasing the amount of control they retain. Automation should be inspectable. State should be understandable. Knowledge should persist. Local work should remain local. Intelligence should strengthen engineering judgment rather than replace it.
The objective is not automation for its own sake.
It is better engineering flow.
Website
https://frictionlab.dev
General
contact@frictionlab.dev
Support
support@frictionlab.dev