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) connect your assistant (hooks handle agent identity automatically)
vybe hook install            # Claude Code
# OR
vybe hook install --opencode # OpenCode

# 3) verify
vybe status --check

If vybe status --check returns query_ok=true, setup is done.

First Task

# create work
vybe task create --request-id task_1 --title "Ship feature" --desc "Implement X"

# get current focus + context
vybe resume --request-id resume_1

# close when done (task ID is returned by create)
vybe task set-status --request-id done_1 --id <TASK_ID> --status completed

If a session crashes, run vybe resume and keep going.

Capabilities

Capability	What it does
Task lifecycle	Create, begin, complete, and block tasks with priority and dependencies
Event log	Append-only log of agent activity — what happened, in order
Scoped memory	Key-value pairs stored per-project, per-task, or globally — with expiration
Resume	Restores the agent's full working context from a single command
Safe retries	Every write accepts a --request-id; sending it twice won't create duplicates
Multi-agent	Multiple agents share the same database safely
Hook integration	One-command install for Claude Code and OpenCode
Project scoping	Group tasks and memory under named projects
Maintenance	Automatic cleanup of old events — configurable in config.yaml
SQLite	Single file, no server, handles concurrent access out of the box

See docs/ for internals: focus selection algorithm, concurrency model, idempotency protocol, event archiving.

Architecture

cmd/vybe/main.go
  ↓
internal/commands/     # CLI layer — parse flags, call actions
  ↓
internal/actions/      # Business logic — orchestrate store calls
  ↓
internal/store/        # SQLite persistence — transactions, retry, conflict resolution

Commands: artifacts, events, hook, loop, memory, push, resume, schema, status, task, upgrade

See docs/ for full documentation.

Documentation

Doc	Contents
docs/operator-guide.md	Install/bootstrap plus operational loop recipes
docs/agent-contract.md	Canonical machine I/O and integration contract
docs/contributor-guide.md	Contributor workflow for safe code changes
docs/decisions.md	Why command-surface choices exist and what must not regress

Uninstall

vybe hook uninstall            # Claude Code
vybe hook uninstall --opencode # OpenCode

State is stored in ~/.config/vybe/. Remove that directory to wipe all data.

License

MIT