Open-source kern: a runtime for LLM agents with a single persistent...

Open-source kern: a runtime for LLM agents with a single persistent session

On Hacker News, Show HN introduces kern (npm package kern-ai), an open-source runtime for personal LLM agents where all interfaces—from terminal to Telegram and IRC—operate on a single persistent session, and memory follows the agent-kernel pattern, collapsing conversations into a hierarchy of L0→L1→L2 summaries.

Compiled by Sergey KostenchukPublished 2026-09-23Updated 2026-09-23

2026-09-23 Coding OpenAI

🛠 kern — an agent runtime with a single persistent session

Show HN introduces kern (npm package kern-ai) by oguzbilgic: TUI, web UI, Telegram, Slack, Matrix, Nostr, and IRC all operate on a single session. Memory is divided into topical segments, collapsed into L0→L1→L2 summaries, and searched via recall. Models: OpenRouter, Anthropic/OpenAI, or Ollama.

🌍 The engineering novelty: agent memory patterns are consolidated into a single runtime—a step toward the “agent as a long-running process” paradigm. The effectiveness of compression remains a hypothesis without measurements, and a bot forwarding other people's messages may run into platform rules.

👤 Install with two commands (npm install -g kern-ai, kern init) or via Docker with a Telegram bot; with Ollama, it runs locally and for free. The agent builds HTML dashboards and spawns read-only sub-agents via spawn.

Source 1: https://github.com/oguzbilgic/kern-ai Source 2: https://github.com/oguzbilgic/kern-ai/blob/master/templates/KERN.md

Sources