Ark

A lightweight, high-performance Slack gateway for AWS Bedrock Agents. Single static binary, under 20 MB memory, connects via Socket Mode (no public endpoint needed).

• Direct messages - the bot responds in a flat conversation
• Channel mentions - the bot responds in a thread

Slack thread timestamps are used as Bedrock session IDs, so follow-up messages in the same thread maintain conversation context.

Prerequisites

• A Slack app with Socket Mode enabled and scopes: app_mentions:read, chat:write, im:history, reactions:write
• An AWS Bedrock Agent with an alias

Configuration

All configuration is via environment variables:

AWS credentials are resolved automatically: explicit keys > ECS task role > AWS CLI (SSO, assume-role, instance profile).

Development

cp .env.example .env  # fill in your values
make build            # debug binary
make test             # run specs
make lint             # run ameba
make format           # format source

Environment variables in your shell take precedence over .env values.

Docker

docker build -t ark .
docker run -e SLACK_BOT_TOKEN=... -e SLACK_APP_TOKEN=... \
  -e BEDROCK_AGENT_ID=... -e BEDROCK_AGENT_ALIAS_ID=... \
  ark

The image uses a multi-stage build (Crystal Alpine -> Alpine) with a static binary. Ark uses ~10 MB of memory and near-zero CPU at idle — it runs comfortably on the smallest instances (e.g., t4g.nano).

Documentation

Full documentation is available at crystal-autobot.github.io/ark.