Vybe

vybe

Persistent memory for AI coding agents.

The Problem

Your agent is halfway through a refactor — six files changed, two to go — and the session dies. Context window limit, crash, timeout, whatever.

Next session starts cold. The agent re-reads files it already understood, re-plans work it already decided on, and redoes changes it already made.

**Agents have no memory that survives a session boundary.** There's nowhere to store what they were doing, what they learned, or what they produced.

`vybe` is that storage layer — a single SQLite file that survives crashes and restores context on restart.


What It Does

On session start, `vybe resume` returns everything the agent needs — its current task, saved memory, recent activity, and linked files.

Between sessions, vybe stores:

• Tasks with status, priority, and dependency tracking
• Events as an append-only log of everything the agent did
• Memory as key-value pairs scoped to global, project, task, or agent — with expiration
• Artifacts linking files and outputs to the work that produced them

Sending the same command twice is safe — duplicates are detected and ignored. Multiple agents share the same database without stepping on each other.

Quick Start

# 1) install
go install github.com/dotcommander/vybe/cmd/vybe@latest

# 2) one-step setup: config dir, database, hooks, default agent
vybe init

# 3) verify
vybe doctor

`vybe init` is idempotent — re-running it is safe and skips already-applied steps. `vybe doctor` confirms the binary is on PATH, hooks are installed, and the database is reachable. Run it any time to validate setup.

First Task

Set your agent identity once — export `VYBE_AGENT` (or set `default_agent` in `~/.config/vybe/config.yaml`) so you never pass `--agent` again:

export VYBE_AGENT=claude

# create work
vybe task create --title "Ship feature" --desc "Implement X"

# get current focus + context
vybe resum