Ankole

What Ankole does

Ankole runs AI agents that do ongoing work for a company, reached in chat tools such as Slack, Microsoft Teams, Lark, DingTalk and WeCom. It is written in Elixir and presents itself as an open source alternative to Claude Tag. It is not tied to Slack or to Claude models, and model and search calls go only to the providers you configure, which can include a model you host.

The Company Brain learns from conversations, registered files and URLs, and notes agents write, storing each item as a claim with source, time, confidence, audience and holder. Conflicting claims stay visible and old versions are kept. Each person and agent has an identity and permissions, and before the model reads company knowledge the brain removes what the asker may not see. Agents are woken by messages, schedules or webhooks, long tasks resume after crashes using state kept in PostgreSQL, and each agent has its own Linux container. Token quotas cap spend per agent. The repository has a hermes-agent topic, but the opening part of the README does not describe a Hermes link.

Key features

• Company Brain storing claims with source, time, confidence and audience
• Per-agent and per-person identity and permissions
• Agents triggered by chat messages, schedules and webhooks
• Long tasks recovered after crashes, with history in PostgreSQL
• A Linux container per agent
• Token quotas per agent and offboarding that stops a person's tasks

When to use it

• Deploying a team assistant in Slack that remembers channel decisions
• Running scheduled agent work with permissions per person
• Keeping company context on your own servers while choosing any model provider

Who it is for: Companies that want self-hosted chat-based AI agents with shared, permissioned memory.

How it fits with Hermes Agent

Tagged hermes-agent, but the opening part of the README does not describe how Ankole relates to Hermes Agent.

FAQ