Last Light

Keep the lights on after you've moved on

An open-source agent that looks after repos in maintenance mode. Triages issues, reviews PRs, chases down stale items — so you can focus on what's next.

Get Started View Source Roadmap

Keeping the lights on

Issue Triage

Labels, deduplicates, and chases missing info — every 15 minutes, rain or shine.

PR Review

Reads the diff, leaves honest feedback. Critical issues first, nits last. No feelings spared.

Health Reports

Monday morning summary of what's piling up, what's gone stale, and what needs you.

Runs Itself

Cron jobs run triage, reviews, and reports on autopilot. Set it up once, then move on.

How it works

Built on Hermes Agent with a custom MCP server packing 28 GitHub tools. Authenticates as its own GitHub App — every action is clearly from the bot, not you.

1
Point it at your repos

List what you want watched in .hermes.md

2
You move on to other things

Cron jobs keep running — scanning for new issues, unreviewed PRs, stale items

3
Last Light gets to work

Labels, reviews, closes duplicates, writes reports

4
The repo stays healthy

Everything happens on GitHub — comments, labels, reviews. No extra dashboard to check

Light it up

Step 1 Install Hermes Agent

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

Step 2 Clone Last Light

git clone https://github.com/cliftonc/lastlight.git
cd lastlight

Step 3 Create a GitHub App

Head to github.com/settings/apps/new and create an app with these permissions:

Generate a private key, save the .pem file into the lastlight folder, and install the app on your repos.

Step 4 Configure

cp .env.example .env
cp config.yaml.example config.yaml

Fill in your GitHub App credentials in .env:

GITHUB_APP_ID=123456
GITHUB_APP_PRIVATE_KEY_PATH=./your-app.private-key.pem
GITHUB_APP_INSTALLATION_ID=789012

Then add your repos to .hermes.md:

## Managed Repositories
- yourname/your-repo
- yourname/another-repo

Step 5 Choose a model

Last Light needs an LLM to think with. Set the provider and model in config.yaml:

model:
  default: anthropic/claude-sonnet-4  # or any model your provider supports
  provider: openrouter                # routes to 100+ models via one key

Then add your API key to .env. Pick whichever provider you prefer:

OpenRouter

One API key, many models. OpenAI, Anthropic, Google, open-source — all through one endpoint.

OPENROUTER_API_KEY=sk-or-...

OpenAI

Direct access to GPT and Codex models.

provider: openai
# or: provider: openai-codex

Anthropic

Direct access to Claude models.

provider: anthropic

AWS Bedrock

Use models via your AWS account. Needs AWS credentials configured.

provider: bedrock

Hugging Face

Open models via inference providers. Free tier included.

HF_TOKEN=hf_...

OpenCode

Curated models (GPT, Claude, Gemini, open-source). Pay-as-you-go.

OPENCODE_ZEN_API_KEY=...

Step 6 Pick a terminal backend

Last Light runs commands through Hermes Agent's terminal. By default it uses local — commands run directly on your machine. For isolation or remote execution, pick a different backend in config.yaml. See the Hermes configuration docs for full details:

Docker

Sandboxed container with dropped capabilities, PID limits, and tmpfs. Needs Docker installed.

terminal:
  backend: docker

Modal

Cloud VM sandbox. Ephemeral, scalable, great for evals. Needs a Modal account.

terminal:
  backend: modal

Set MODAL_TOKEN_ID and MODAL_TOKEN_SECRET in .env.

SSH

Run on a remote server. Persistent shell with connection reuse. Good for beefy hardware.

terminal:
  backend: ssh

Set TERMINAL_SSH_HOST and TERMINAL_SSH_USER in .env.

Daytona

Managed cloud dev environment with stop/resume. Needs a Daytona API key.

terminal:
  backend: daytona

Singularity

For HPC clusters and shared machines where Docker isn't available. Uses Apptainer/Singularity.

terminal:
  backend: singularity

Local not recommended

Commands run directly on your machine. No isolation — the agent has full access to your filesystem.

terminal:
  backend: local

All container backends support resource limits:

terminal:
  container_cpu: 1         # CPU cores
  container_memory: 5120   # Memory in MB
  container_disk: 51200    # Disk in MB
  container_persistent: true

Step 7 Connect a gateway optional

Talk to Last Light from Discord, Slack, Telegram, and more. See the Hermes gateway docs for the full list. Run the setup wizard:

./lastlight gateway setup

Then run it as a persistent service:

./lastlight gateway            # run in foreground
./lastlight gateway install    # install as a system service

Discord

Create a bot at the Discord Developer Portal. Enable Presence Intent and Message Content Intent under Privileged Gateway Intents or the bot won't connect.

DISCORD_BOT_TOKEN=your-token
DISCORD_HOME_CHANNEL=channel-id

Slack

Create an app at api.slack.com/apps with bot token scopes, then set in .env:

SLACK_BOT_TOKEN=xoxb-...
SLACK_APP_TOKEN=xapp-...

Telegram

Create a bot via @BotFather, then set in .env:

TELEGRAM_BOT_TOKEN=your-token
TELEGRAM_HOME_CHANNEL=chat-id

GitHub Webhooks

React to repo events in real time. Enable in .env and point your GitHub App's webhook URL at it. The endpoint must be publicly reachable — use a tunnel like ngrok or Cloudflare Tunnel if running locally.

WEBHOOK_ENABLED=true
WEBHOOK_PORT=8644
WEBHOOK_SECRET=your-secret

Step 8 Run it

./lastlight

That's it. You can also run single commands:

./lastlight chat -q "Review the latest PR on yourname/your-repo"
./lastlight chat -q "Triage open issues on yourname/another-repo"

Tricks up its sleeve

/pr-review

Review a PR with structured, severity-ranked feedback

/issue-triage

Label, deduplicate, and prioritize issues

/repo-health

Generate a health report with metrics and action items

Want more?

Check the roadmap for what's coming next

On autopilot

These run automatically, out of the box:

Job Frequency
Triage new issues Every 15 min
Check PRs awaiting review Every 30 min
Weekly health report Mondays 9am
More coming soon... Roadmap