README
¶
Git for AI Agents
Version control for AI agent activity. Track what your agent did, which prompt wrote each line, and rewind when things break.
Demo
Every tool call is automatically captured. No manual commits needed.
Quick Start
# Install
go install github.com/regent-vcs/regent/cmd/rgt@latest
# Initialize in your project
cd your-project
rgt init
# Work with Claude Code normally (every tool call is tracked)
# See what happened
rgt log
That's it. Your agent activity is now auditable.
What You Get
See what your agent actually did
$ rgt log
Step a1b2c3d | 2 min ago | Tool: Edit
│ File: src/handler.go
│ Added error handling to request handler
│ + 5 lines, - 2 lines
Step d4e5f6g | 5 min ago | Tool: Write
│ File: tests/handler_test.go
│ Created unit tests for handler
│ + 23 lines
Step f8g9h0i | 8 min ago | Tool: Bash
│ Command: go mod tidy
│ Cleaned up dependencies
Blame: which prompt wrote this line?
$ rgt blame src/handler.go:42
Line 42: func handleRequest(w http.ResponseWriter, r *http.Request) {
Step: a1b2c3d4e5f6
Session: claude-20260502-143021
Tool: Edit
Prompt: "Add error handling to the request handler"
Track multiple concurrent sessions
$ rgt sessions
Active Sessions:
claude-20260502-143021 | 3 steps | Last: 2 min ago
claude-20260502-091534 | 7 steps | Last: 2 hours ago
$ rgt log --session claude-20260502-143021
# Filter history by session
See full context for any change
$ rgt show a1b2c3d
Step a1b2c3d4e5f6
Parent: d4e5f6g7h8i9
Session: claude-20260502-143021
Time: 2026-05-02 14:30:21
Tool: Edit
File: src/handler.go
Changes:
+ func handleRequest(w http.ResponseWriter, r *http.Request) {
+ if r.Method != "GET" {
+ http.Error(w, "Method not allowed", 405)
+ return
+ }
- func handleRequest(w http.ResponseWriter, r *http.Request) {
Conversation:
User: "Add error handling to reject non-GET requests"
Assistant: "I'll add method validation to the handler..."
Why This Exists
The problem: AI agents have no version control of their own.
You know this pain:
- "It was working five minutes ago"
- "Why did you change that file?"
- "Go back to before the refactor"
/compactand pray- Copy-pasting code into a fresh chat
The solution: Three primitives that should already exist:
rgt log— what did this session do?rgt blame— which prompt wrote this line?rgt rewind— restore to any previous step (coming soon)
We gave agents write access to our codebases. We did not give ourselves git for it. Regent fixes that.
How It Works
Regent stores agent activity in .regent/ (like .git/):
.regent/
├── objects/ # Content-addressed blobs (BLAKE3)
├── refs/ # Session pointers (one per agent)
├── index.db # SQLite query index
└── config.toml
Every tool call creates a Step:
Step {
parent: <previous-step-hash>
tree: <workspace-snapshot>
transcript: <conversation-delta>
cause: {
tool_name: "Edit"
args: <what-changed>
result: <tool-output>
}
session_id: "claude-20260502-143021"
timestamp: "2026-05-02T14:30:21Z"
}
Steps form a DAG. Each session has its own branch. Common ancestors dedupe. You get git-level auditability for agent activity.
Technical details: See POC.md for the complete specification.
Installation
Via Go Install (Recommended)
go install github.com/regent-vcs/regent/cmd/rgt@latest
From Source
git clone https://github.com/regent-vcs/regent
cd regent
go build -o rgt ./cmd/rgt
sudo mv rgt /usr/local/bin/
Coming Soon
- Homebrew:
brew install regent-vcs/tap/rgt - Binary releases: GitHub Releases
Commands
Available Now:
| Command | Description |
|---|---|
rgt init |
Initialize .regent/ in current directory |
rgt log |
Show step history (supports --session, -n, --since) |
rgt sessions |
List all active sessions |
rgt status |
Show current repository state |
rgt show <step> |
Display full context for a step |
rgt cat <hash> |
Inspect any object by hash (debug) |
Coming Soon:
| Command | Status | Description |
|---|---|---|
rgt blame <path>:<line> |
Phase 3 | Per-line provenance with prompt |
rgt rewind <step> |
Phase 5 | Non-destructive time-travel |
rgt gc |
Phase 6 | Garbage collection |
Features
- Content-Addressed Storage — BLAKE3 hashing, automatic deduplication
- Fast Queries — SQLite index, sub-10ms lookups
- Per-Session DAG — Concurrent agents, no conflicts
- Conversation Tracking — Survives
/compactand/clear - Hook-Driven — Transparent Claude Code integration
- Concurrency-Safe — CAS refs, ACID transactions
- Gitignore-Compatible —
.regentignoresupport
Regent vs Git
| Git | Regent | |
|---|---|---|
| Tracks code | ✅ | ✅ |
| Tracks agent activity | ❌ | ✅ |
| Blame with prompt | ❌ | ✅ |
| Conversation history | ❌ | ✅ |
| Concurrent sessions | ⚠️ conflicts | ✅ separate branches |
| Purpose | Developer VCS | Agent audit trail |
Regent complements git, doesn't replace it. Use both.
Roadmap
- Phase 1: Object store (blob, tree, step, ref) — COMPLETE
- Phase 2: Hook integration (Claude Code) — COMPLETE
- Phase 3: Blame algorithm (Myers diff) — IN PROGRESS
- Phase 4: Transcript capture (JSONL) — PLANNED
- Phase 5: Rewind (time-travel) — PLANNED
- Phase 6: Concurrency hardening — PLANNED
Check GitHub Projects for current priorities.
Status
Active Development (POC Stage)
- ~4.5k LOC Go implementation
- Core functionality works (Phases 1-2 complete)
- Used in production by contributors
- Not yet v1.0 (see roadmap)
Honest assessment: Production-quality code at POC-level feature completeness. We're building in public.
Contributing
We use GitHub Flow. Create a branch, add commits, open a PR.
Read CONTRIBUTING.md for details.
Before opening a PR:
- Tests pass:
go test ./... - Code formatted:
gofmt -w . - Read BRAND.md if touching CLI output
Built With
- cobra — CLI framework
- blake3 — BLAKE3 hashing
- go-diff — Myers diff
- modernc.org/sqlite — Pure Go SQLite
License
Built by contributors
Directories