vela-cli 0.1.1

Runtime observability CLI for AI agents

Vela CLI

Runtime observability for AI agents — no code changes required.

Install

# Homebrew (macOS / Linux) — recommended
brew tap vela-ai/vela
brew install vela

# pip — use a virtual environment
python3 -m venv venv
source venv/bin/activate
pip3 install vela-cli

Virtual environments are required on Python 3.11+ due to system protection. On Windows, use venv\Scripts\activate instead.

Quick start

vela login
vela wrap python my_agent.py

That's it. Every LLM call is captured and visible at vela.wtf/dashboard.

---

Commands

vela login

Opens vela.wtf/cli-auth in your browser. After authenticating, your API key is saved to ~/.vela/config.json. Takes about 10 seconds.

$ vela login
Opening browser...
Waiting for authentication...
✓ Logged in as [e-mail]. API key saved.

vela logout

Removes ~/.vela/config.json.

$ vela logout
✓ Logged out.

vela wrap <command>

Runs any command with transparent LLM interception. Every request to OpenAI, Anthropic, Gemini, Groq, or OpenRouter is captured — model, latency, tokens, cost — and streamed to your Vela dashboard in real time.

$ vela wrap python my_agent.py
[vela] Session started · a3f2bc91
[vela] gpt-4o · 1.2s · $0.0043
[vela] claude-3-5-sonnet-20241022 · 0.8s · $0.0031
[vela] Session complete · $0.0074 · view: https://vela.wtf/session/a3f2bc91...

Works with any language or runtime:

vela wrap python my_agent.py
vela wrap node agent.js
vela wrap python -m my_module --flag value

How it works: vela wrap starts a local MITM proxy (port 9877) and sets HTTP_PROXY / HTTPS_PROXY for the subprocess. The proxy intercepts LLM API responses, extracts trace data, and POSTs it to the ingest API without affecting the wrapped process at all.

SSL: The mitmproxy CA certificate (~/.mitmproxy/mitmproxy-ca-cert.pem) is set via SSL_CERT_FILE / REQUESTS_CA_BUNDLE / NODE_EXTRA_CA_CERTS so wrapped processes trust the proxy without system-wide cert changes.

vela status

Shows your account and the last 5 sessions.

$ vela status
Logged in as [e-mail]

  a3f2bc91  complete  1.2s  $0.017
  b7e1da02  complete  1.4s  $0.019
  c8f3ea14  error     0.8s  $0.012
  d4a9fb27  complete  1.1s  $0.016
  e2b5gc38  running   —     —

vela init

Detects your project's entry file and inserts the Vela SDK initialisation block at the top. Asks for confirmation before writing.

$ vela init
Detected: my_agent.py (OpenAI, LangChain)

Will add to top of my_agent.py:

  import vela
  vela.init(api_key="vela_xxx")

Add Vela to my_agent.py? [Y/n]: y
✓ Added Vela to my_agent.py

vela update

Upgrades to the latest release.

$ vela update
✓ vela-cli updated.

---

Supported LLM providers

Provider	Host intercepted
OpenAI	api.openai.com
Anthropic	api.anthropic.com
Google Gemini	generativelanguage.googleapis.com
Groq	api.groq.com
OpenRouter	openrouter.ai

---

Offline mode

If vela.wtf is unreachable, vela wrap still runs the wrapped process normally and prints:

[vela] Offline mode — traces will not be saved.

The process is never interrupted for network reasons.

---

Homebrew tap

The tap lives at github.com/vela-ai/vela. The formula at Formula/vela.rb in this directory is the canonical source.

brew tap vela-ai/vela
brew install vela

---

Requirements

• Python 3.9+
• mitmproxy (installed automatically as a dependency)