CLI
vectorgraph the first party command line for scripts and CI
Scoped tokens, structured JSON, and explicit idempotency keys. The release-candidate CLI covers the supported workspace operations scripts and CI need: issues, comments, context bundles, deploy markers, and NorthGraph workflows under the same workspace and private-team rules.
$ vectorgraph north diff analyze --base main --json
"verdict": "violations",
"risk": "high",
"summary": { "filesChanged": 14, "routesImpacted": 1 }
Release package path
Install, log in, and connect your agents
The current public package is @orintech/[email protected]. Verify that exact registry release before installing it. The same package provides the workspace CLI, NorthGraph, and local NorthGraph MCP; browser login starts reviewed hosted MCP setup for detected Codex and Claude clients when the user approves it. Its non-mutating workspace selector and saved-profile session routing let parallel processes target separate workspaces without changing the default. The complete current sequence is in the get-started guide. Version 0.7.0 includes the exact team UUID selection correction used by scoped multi-workspace agent setup and the standard vectorgraph --version alias. Both paired packages are published and registry-verified at that exact version.
- 01
Verify and install the exact release
npm view @orintech/[email protected] version npm install --global @orintech/[email protected] - 02
Log in and verify the active workspace
vectorgraph auth login vectorgraph auth status - 03
Configure Codex and Claude for the active workspace
vectorgraph workspace use <workspace-id-or-slug> vectorgraph mcp setup --client codex,claude --team ENG --local-north
The commands teams reach for first
Six commands cover the work most automations need
Everything else the binary supports follows the same conventions: JSON output, structured errors, scoped tokens, and explicit idempotency keys for retryable mutations. Run vectorgraph --help for the full surface.
vectorgraph issues queryRead and filter issues. Pipe straight into jq, a Slack post, or a CSV export. Stable schema for CI scripts.
vectorgraph issues createOpen issues from any automation with full field support. Idempotency keys mean retries don't double-file.
vectorgraph signals submitCapture customer signal from support pipelines, analytics anomaly handlers, or sales call notes.
vectorgraph deploy markTag a deploy. Auto-closes the issues in the release and notifies the originating signal requesters.
vectorgraph context bundleFetch the same permission-filtered context bundle agents see through MCP. Useful for hybrid scripts.
vectorgraph north diff analyzeRun NorthGraph against a diff locally. Same verdict your CI gate will see, returned as structured JSON.
Release contract example
JSON output you can pipe directly into the next command
The example uses supported flags and illustrates the structured release contract. Verify the installed package and target environment before depending on an exact response shape.
# Find every started ENG issue$ vectorgraph issues query --team eng --status started --json \ | jq -r '.issues[] | .identifier'ENG-2189ENG-2204ENG-2211 # Mark a deploy and auto-close the issues it shipped$ vectorgraph deploy mark --issues ENG-2189 --sha c93af1e --env production --json{ "deploy": { "environment": "production", "sha": "c93af1e" }, "issues": [ { "issueIdentifier": "ENG-2189", "statusCategory": "completed" } ], "requestId": "req_..."}In your pipeline
Drop into any CI in two lines
Use the official GitHub Action after the release workflow publishes it, or call the binary directly from any shell. Stable exit codes mean a failed NorthGraph check fails the build, no parsing required.
# .github/workflows/north-graph.ymlname: NorthGraph analyzeon: pull_request: jobs: analyze: runs-on: ubuntu-latest steps: - uses: actions/[email protected] - uses: vectorgraph/setup-vectorgraph@v1 with: token: ${{ secrets.VECTORGRAPH_TOKEN }} - run: vectorgraph north diff analyze --base origin/main --jsonGet started
Bring your team onto the shared work graph
Create an account, name a workspace, invite your team, connect your code, and point your agents at MCP. Start turning company context into shipped work from one shared workspace.