v3 — agent CLI for renketsu

Terminal client for renketsu, a context-graph engine for AI agents.

Install

npm install -g @codespring-app/v3

Requires Node.js 22+.

Quick start

# 1. Log in via the OAuth device flow (opens a browser tab)
v3 auth login

# 2. List your projects
v3 project list

# 3. Pick one as the default for subsequent commands
v3 project use proj_xxxxxxxx

# 4. Search the project's context graph
v3 search "what was decided about the auth flow"

# 5. Ask a question (GraphRAG global/local)
v3 ask "what did Alice work on last week" --mode auto

# 6. Stream a long-running ingest
v3 ingest --file ./notes.md --format extract

Auth

v3 auth login runs the standard RFC 8628 device-authorization flow:

1. The CLI requests a device code.
2. It opens https://renketsu.codespring.app/device and prefills the user code.
3. You approve in the browser.
4. The CLI polls until the API hands back a session token, which is stored at ~/.config/v3/credentials.json (mode 0600).

Override the API URL for self-hosted instances:

v3 auth login --api-url https://your-instance.example.com
# or via env
V3_API_URL=https://your-instance.example.com v3 ...

Commands

v3 auth login | logout | whoami

v3 project list | get <id> | use <id> | create --name NAME

v3 search "<query>" [--top-k N] [--project ID]
v3 stats [--project ID]
v3 graph [--limit N] [--project ID]

v3 ingest --file PATH --format extract|structured|raw [--project ID]
v3 ask "<question>" [--mode auto|global|local] [--project ID]

v3 job list | get <id> | cancel <id> | tail <id>

Run v3 --help for the full command surface.

License

MIT.