Local-first tools for
AI coding agents
AI coding agents are powerful but stateless, uncoordinated, and token-wasteful. Every other tool vendor is building cloud-hosted solutions that require sending your code context, decisions, and often secrets to their servers.
The bye suite takes the opposite approach. Everything runs on your machine. Nothing phones home. Your data never leaves your control.
The Tools
Your AI agent forgets everything between sessions. byecontext fixes that. It gives Claude Code (or any MCP-compatible agent) persistent memory — decisions, errors, solutions, and architecture notes that survive across sessions. PostgreSQL for structured storage, Vespa for vector search over your memory.
Run multiple AI coding sessions in parallel without them stepping on each other. byeipc provides a work queue and session management layer so agents can send tasks, claim work, and coordinate — all through a Go CLI backed by PostgreSQL on localhost. No cloud broker. No vendor lock-in.
LLMs waste tokens parsing verbose command output. byetoken intercepts Bash commands via a hook, runs them through semantic compression, and returns output that's 99.7% smaller while retaining 100% of the information the LLM needs. Built on ripgrep's stack. Self-tuning. The LLM never knows it's there.
Why Local-First
Your data stays yours
Session memory, code context, architectural decisions, error logs — none of it leaves your machine. No cloud accounts. No telemetry. No “anonymous” data collection.
Works offline
Every tool runs against local storage — PostgreSQL on localhost, Vespa, the filesystem. Network goes down? Your agent workflow keeps working.
No vendor lock-in
Apache-2.0 licensed. Standard protocols (MCP, stdin/stdout). Works with Claude Code today. Will work with whatever replaces it tomorrow.