Skip to content

What Skylattice Is

For the canonical search-friendly overview page, start with what-is-skylattice.md. For the current stable release signal, start with releases/v0-3-0.md.

Skylattice is a local-first AI agent runtime for people who want an agent they can inspect like software.

Most agent projects optimize for breadth, speed, or workflow convenience. Skylattice starts from a different question:

What if a personal agent had to keep durable memory, act through explicit governance, and evolve through reviewable Git changes instead of hidden behavior drift?

That question shapes the whole repository.

The Core Idea

Skylattice splits agent state into two surfaces on purpose:

  • private runtime state lives under .local/
  • durable system behavior lives in tracked files such as docs, configs, prompts, ADRs, and tests

This makes the runtime easier to reason about:

  • memory can be persistent without becoming public by accident
  • behavior changes can be reviewed like code
  • automation can be bounded by policy instead of implied by prompts
  • self-improvement can be rollbackable instead of mystical

What It Already Demonstrates

Today the repository shows two concrete workflows.

1. Task Agent

A governed task path that can:

  • plan bounded repo work
  • materialize deterministic text edits
  • validate changes with tracked commands
  • record ledger events and memory writes
  • prepare a draft PR when external writes are allowed

2. Technology Radar

A bounded discovery path that can:

  • scan GitHub repositories through tracked topics and scoring rules
  • record evidence and candidate scores
  • run limited experiments on whitelisted paths
  • promote tracked changes through a rollbackable Git path

Why The Design Matters

Skylattice is useful even if you never run it in production, because it is a compact reference for several hard problems:

  • how to separate private agent memory from public repository state
  • how to make agent actions auditable after the fact
  • how to keep CI and runtime validation aligned
  • how to bound self-modification without pretending it does not exist

Who It Is For

Skylattice is a strong fit for:

  • builders exploring local-first agent infrastructure
  • contributors interested in governance-heavy automation
  • people who want a reference repo for memory, ledger, and Git-backed review boundaries

It is not yet a strong fit for:

  • users expecting a hosted product
  • teams wanting broad integrations out of the box
  • workflows that need unrestricted tool use or autonomous production operations

Where To Go Next

  • start with index.md for the canonical landing page and quickstart.md for the proof-first walkthrough
  • read use-cases.md for concrete user-facing scenarios
  • read comparison.md to understand what Skylattice is and is not competing with
  • read architecture.md if you want the runtime map
  • open a GitHub issue with early feedback on the public repository