Laminar Agent

What it is

Laminar Agent fits the bring-your-own-everything story: you already deploy to your own AWS, now you bring your own model too. Point it at any OpenAI-compatible LLM and it works against your project’s deploy state and CloudWatch logs — no data leaves for a third-party SaaS beyond the LLM provider you choose. It runs from the same CLI you already use to deploy.

There are three things it does:

Bring your own key

The agent talks to any OpenAI-compatible endpoint — OpenAI itself, OpenRouter, or a local model server (Ollama, vLLM, LM Studio). Configure it with environment variables:

export LAMINAR_AGENT_API_KEY=sk-...                  # your provider key
export LAMINAR_AGENT_BASE_URL=https://api.openai.com/v1   # default; any OpenAI-compatible provider
export LAMINAR_AGENT_MODEL=gpt-4o-mini               # any model the provider exposes

LAMINAR_AGENT_BASE_URL defaults to https://api.openai.com/v1. Override it to use another provider:

# OpenRouter
export LAMINAR_AGENT_BASE_URL=https://openrouter.ai/api/v1
export LAMINAR_AGENT_MODEL=anthropic/claude-3.5-sonnet

# Local model (Ollama)
export LAMINAR_AGENT_BASE_URL=http://localhost:11434/v1
export LAMINAR_AGENT_MODEL=llama3.1

You can also configure it per-project with an agent: block in laminar.yml (env vars take precedence so you can keep keys out of source control):

agent:
  base_url: https://api.openai.com/v1
  model: gpt-4o-mini
  # api_key is best left in the env (LAMINAR_AGENT_API_KEY)
  slack_webhook_url: ${SLACK_WEBHOOK_URL}

Diagnose a failed deploy

Pass a stage and the agent pulls the latest deployment status plus recent logs, then summarizes the root cause and what to do next.

$ laminar agent diagnose production
› Root cause
  Lambda init timeout: DB connection to rds-prod blocked — the function's
  security group has no egress to the database subnet.
› Next step
  Add an egress rule on the SG (or attach the RDS SG) and redeploy.

Add --slack to also post the summary to your Slack channel.

Ask questions

Ask anything about your deploy state and logs in plain English. Scope it to a stage with --stage.

laminar agent ask "why did production roll back last night?"
laminar agent ask "what's the cold-start p95 right now?" --stage staging

Watch for failures

Run watch to poll a stage for failures and get alerted the moment something breaks — useful right after a deploy or in a long-running session.

laminar agent watch production --slack

Slack setup

The --slack flag posts to Slack via an incoming webhook. Create one for the channel you want, then expose it as SLACK_WEBHOOK_URL (or set slack_webhook_url in the agent: block).

export SLACK_WEBHOOK_URL=https://hooks.slack.com/services/T000/B000/XXXX
laminar agent diagnose production --slack   # → posts the summary to that channel

Without --slack the agent only prints to your terminal — Slack is fully opt-in.

Telegram setup

Prefer Telegram? Bring your own bot and the agent will send diagnoses straight to a chat you control — no shared third-party app, the bot is yours. It works alongside Slack and any other notification channel you’ve configured.

First, create a bot: message @BotFather on Telegram, run /newbot, and it will hand you a bot_token. Then find your chat_id: send your new bot any message, then open https://api.telegram.org/bot<bot_token>/getUpdates and read the chat.id from the response.

Add both to the agent: block of laminar.yml (or set the TELEGRAM_BOT_TOKEN / TELEGRAM_CHAT_ID env vars, which take precedence):

agent:
  base_url: https://api.openai.com/v1
  model: gpt-4o-mini
  telegram_bot_token: ${TELEGRAM_BOT_TOKEN}   # from @BotFather
  telegram_chat_id: ${TELEGRAM_CHAT_ID}       # the chat to post diagnoses to

Cloud users don’t edit laminar.yml — set the bot token and chat ID in the agent settings of your dashboard, where they’re stored encrypted.