agent-core 0.1.0

First public release of agent-core: an engine for AI team agents that work alongside people.

What's in it

• Agent engine: turns, model tiers on Bedrock or OpenRouter, a tool gateway with allow-lists, per-agent budgets with a hard stop, a kill switch, a scheduler and long-term memory.
• Channels: Teams, email, phone and voice, Jira, GitHub.
• Documents, plans and an organisation-neutral skill pack.
• Helm charts for always-on agents.
• A wake-on-demand trial (deploy/scale-to-zero-poc): a front door that holds the channels, a warm pool, workers in OpenShell sandboxes that start on a message and exit when idle, and a memory proxy in front of Hindsight. It's a trial, not a production setup: read SECURITY.md first.

Image: docker pull ghcr.io/sinkers/agent-core:0.1.0

See the README to get started, and SECURITY.md for known gaps and how to report a vulnerability privately. MIT licence.