BurnOS 0.27.0

Memory server on Mnemosyne 1.7.0: books become BurnSkills in full, built in the background on the working model; a codex: model builds on the ChatGPT subscription (codex:gpt-6.1-sol in Settings). Update zip: extract over an existing install.

BurnOS v0.24.0 - run on your Claude or ChatGPT subscription

Choose claude:sonnet, claude:opus or codex:default (or type claude:<model> / codex:<model>) as the talking or working model in Settings. Burnos then answers through the official Claude Code or Codex CLI you are logged in to, using your subscription instead of OpenRouter credit.

• Your logins stay with the CLIs. Burnos never reads their login tokens.
• The CLI only writes Burnos's reply. Its own tools are switched off. Every action still goes through Burnos's tools and the confirm-before-acting guard.
• Screenshots still reach the model, so tasks can check their work.
• No OpenRouter key is needed when both models are on a subscription. Memory search by meaning and skill conversion still use OpenRouter when a key is set.
• Speed: measured on this PC, Claude answers each turn in 2 to 4 seconds and Codex in 5 to 7.

Bundles Mnemosyne 1.4.1. The release passed verify-memory.py, verify-windows-package.py and verify-bundled-models.py. Checksums are in SHA256SUMS.txt.