Skip to content

QUICK_START.md

Latest commit

 

History

History
85 lines (60 loc) · 2.91 KB

QUICK_START.md

File metadata and controls

85 lines (60 loc) · 2.91 KB
Raw

Quick start

Install → first merged branch, step by step.

1 · Install + MongoDB

brew install argon-lab/tap/argonctl      # or: npm install -g argonctl

# MongoDB must be a replica set (change streams); one node is fine:
docker run -d --name argon-mongo -p 27017:27017 mongo:7 --replSet rs0
docker exec argon-mongo mongosh --quiet --eval 'rs.initiate()'

Argon connects to MONGODB_URI (default mongodb://localhost:27017) and keeps its metadata in the argon_wal database.

2 · Bring your data in

argon import database --uri mongodb://localhost:27017 --database myapp --project myapp
# starting fresh instead: argon projects create myapp

Every document becomes versioned history on main; a snapshot is taken automatically so reads never replay the whole import.

3 · Branch and work with any driver

argon branches create experiment -p myapp   # instant — a pointer, no copy
argon checkout -p myapp -b experiment       # → prints a MongoDB URI
argon watch    -p myapp -b experiment       # captures writes (keep running)

Point pymongo, mongoose, mongosh — anything — at the URI and work normally: real indexes, aggregation, transactions. Every write becomes history, attributed to its actor.

Want stable URIs instead of per-checkout ones? argon proxy serves mongodb://host:27018/<project>~<branch>?directConnection=true. Prefer a UI? argon console opens a local web console.

4 · Review and merge — a data pull request

argon diff          -p myapp -b experiment   # what would change
argon merge preview -p myapp -b experiment   # persist a reviewable plan
argon merge apply <plan-id>                  # exactly-once; stale heads refused

Conflicts fail loudly; resolve with --strategy theirs|ours or fix the data and re-preview.

5 · Time travel, undo, rewind

argon time-travel query -p myapp -b main --lsn 1000

argon undo -p myapp -b main --from-lsn 990 --dry-run          # revert a range
argon undo -p myapp -b main --from-lsn 990 --actor agent-7    # …or one writer

argon restore preview -p myapp -b main --time 2026-07-07T09:00:00Z
argon restore reset   -p myapp -b main --time 2026-07-07T09:00:00Z --backup pre-incident

Resets are recorded, not destructive: discarded entries stay for audit, and the backup branch (or any pin) keeps the old state readable.

6 · For agents

claude mcp add argon -- argon mcp           # sandboxes/diff/merge/undo/pins as MCP tools
argon sandbox create -p myapp --ttl 1h      # disposable branch + URI, one step
argon pin create  -p myapp --name eval-v1   # immutable dataset state
argon pin sandbox -p myapp --name eval-v1   # identical input, every eval run

Full agent workflow (MCP, REST, Python): AGENTS.md.

Where next

CLI.md — every command · ARCHITECTURE.md — guarantees, honestly · OPERATIONS.md — S3 snapshots, GC, migration