CLI Quick Start

The SWEny CLI is how you build, run, and iterate on AI workflows from your terminal. Describe what you want done in plain English, and SWEny generates a full DAG with the right tools wired up at each step.

Install

npm install -g @sweny-ai/core

The binary is called sweny. You can also run it directly with npx:

npx @sweny-ai/core --help

Build your first workflow

1. Add your API key to .env:

# .env (gitignored)
ANTHROPIC_API_KEY=sk-ant-...

The CLI auto-loads .env at startup — no external dotenv tooling required.

2. Create a workflow from a description:

sweny workflow create "analyze our codebase for common security \
  anti-patterns, compile a report, and create tickets for critical findings"

SWEny generates a full workflow with nodes, conditional routing, and structured output schemas. It renders an ASCII DAG and prompts you to save, refine, or discard:

  Save to .sweny/workflows/security_audit.yml? [Y/n/refine]

Type a refinement instruction to iterate on the design, or press Y to save.

3. Run it:

sweny workflow run .sweny/workflows/security_audit.yml

4. Refine it later:

sweny workflow edit .sweny/workflows/security_audit.yml \
  "add a quality gate that rejects vague findings"

That’s it. No YAML to write by hand, no boilerplate to configure.

Try the built-in triage workflow

SWEny also ships production-ready workflows for SRE triage. No external services needed for a quick test — just an API key and a log file.

1. Create a config file:

sweny new

This runs the interactive picker and writes a starter .sweny.yml. Uncomment the local-only block:

observability-provider: file
log-file: ./sample-errors.json
issue-tracker-provider: file
source-control-provider: file
notification-provider: file

2. Run a dry-run triage:

sweny triage --dry-run

SWEny reads .sweny.yml, loads secrets from .env, analyzes your logs, and writes results to .sweny/output/ without creating any real issues or PRs.

Live progress display

When running on an interactive terminal, the CLI renders a live progress display that updates in place as each node executes:

  ⠹ [2/5] Investigate errors  1m 42s
    ↳ Calling sentry.list_issues
    ↳ Calling github.search_code
    ↳ Analyzing stack traces

The progress block shows a spinner, step counter, node name, and elapsed time. Below it, the last few activity messages stream in as the agent works. When a node finishes, the block is replaced with a completion line:

  ✓ [1/5] Fetch errors  12s
    ↳ Found 3 unresolved issues
  ✓ [2/5] Investigate errors  2m 14s
    ↳ Root cause identified in webhook handler
  ⠹ [3/5] Create issues  4s

Non-TTY environments (CI, pipes) get simple periodic status lines instead of animated output.

NDJSON streaming

Pass --stream to emit raw ExecutionEvent objects as newline-delimited JSON to stdout. This is how Studio’s Live Mode connects to a running workflow:

sweny triage --stream | studio-consumer

--stream can be combined with the normal progress display — structured events go to stdout while the spinner renders to stderr.

Config priority

Settings resolve in this order (highest wins):

CLI flag  >  environment variable  >  .sweny.yml  >  default

Use flags for one-off overrides without editing your config file:

sweny triage --dry-run --time-range 1h --service-filter 'billing-*'

Config file reference

.sweny.yml uses flat kebab-case keys that map 1:1 to CLI flags. You can copy any flag from --help directly into the YAML:

# .sweny.yml -- commit this file. Secrets go in .env.

# Providers
observability-provider: datadog
issue-tracker-provider: github-issues
source-control-provider: github
coding-agent-provider: claude
notification-provider: console

# Investigation
time-range: 24h
severity-focus: errors
service-filter: "*"
investigation-depth: standard

# PR / branch
base-branch: main
pr-labels: agent,triage,needs-review

# Paths
service-map-path: .github/service-map.yml
log-file: ./logs/errors.json

# Cache
cache-dir: .sweny/cache
cache-ttl: 86400

Secrets (API keys, tokens) are never read from the config file — always use environment variables or .env.

Connecting providers

Once local-only mode works, swap in real providers. Set the provider in .sweny.yml and add credentials to .env:

observability-provider: sentry
sentry-org: my-org
sentry-project: my-project
issue-tracker-provider: linear
linear-team-id: your-team-uuid

SENTRY_AUTH_TOKEN=sntrys_...
LINEAR_API_KEY=lin_api_...
GITHUB_TOKEN=ghp_...

Verify credentials before running a full workflow:

sweny check

See the Commands Reference for all supported providers and their options.

What’s next?

E2E Testing — generate and run AI-driven browser tests
Commands Reference — every command, flag, and option
Examples — common configurations and real-world workflows