Changed
OmO Native moves from senpi 2026.9.29-4 to senpi 2026.9.29-5, and the engine changes below come with it.
GPT-6.1 Sol arrives, and deep-low and Geeky · Normal lead with it at medium. (#9214, senpi#2390) The engine adds gpt-6.1-sol and its Fast tier gpt-6.1-sol-fast on the OpenAI API, ChatGPT subscription, Azure, GitHub Copilot, OpenCode Zen, OpenRouter, Venice and Vercel lanes with the documented low..max efforts, and makes it the implicit default for the OpenAI and ChatGPT-subscription providers where GPT-6 Sol was. It runs on the GPT-6 system prompt, which gains the two writing rules from OpenAI's own GPT-6.1 Sol template. The default deep delegation lane now starts on gpt-6.1-sol (medium) from your ChatGPT subscription or the OpenAI API, then gpt-6.1-sol-fast, and keeps GPT-5.6 Sol (medium) and GPT-5.6 Sol Fast behind them, so GitHub Copilot, OpenCode Zen and an engine that does not list GPT-6.1 Sol yet still run the lane on GPT-5.6 Sol as before. The geeky-normal model profile likewise applies GPT-6.1 Sol (medium) first and GPT-5.6 Sol (medium) when 6.1 Sol is not served. The effort stays medium on every rung. The other categories and model profiles are unchanged.
A host can give each session the chat-style app prompt. (senpi#2377) With SENPI_PROMPT_SURFACE=app, or open_session.promptSurface: "app" on a multi-session host, replies drop the terminal routing line and keep tool and hook feedback with the agent unless it changes what you get. Terminal sessions keep today's prompts.
Fixed
Editing a file outside a project no longer asks you to install a language server. (#9223) After an edit to OmO's own config, a scratch file in a temp folder, or any file with no project around it, the LSP check told the agent to install the missing server "in THIS repository" and to ask you first, which would have created a package.json in your home or temp folder, and it came back on every edit. Those files now get no install prompt, and inside a project the prompt for a missing server appears once per session instead of after every edit. Thanks to @haamsuk-collab for the report.
Kibitzer recall and quick delegation work with a single Z.ai or Xiaomi login. (#9202) The quick category had no Z.ai or Xiaomi model, so on a machine logged in to only one of them recall stayed off and quick tasks were unavailable. It now falls back to glm-5.3-flash (Z.ai) and mimo-v2.6-flash (Xiaomi) at low effort, after every model it already used, so machines that had a quick model keep the same one.
Engine fixes from senpi 2026.9.29-5. Remote compaction works again on the ChatGPT subscription lane instead of silently falling back to a local summary (senpi#2378). A session no longer leaves an empty .omo/ folder in the project (senpi#2386). RPC get_auth_providers reports each login method's status separately (senpi#2384). The embedded tree-sitter grammars load on Node again (senpi#2032).
- d131705 Merge pull request #9231 from code-yeongyu/release/v5.1.4-source-state
- 2d26794 Merge pull request #9230 from code-yeongyu/docs/changelog-5.1.4
- 505439e docs(changelog): order the 5.1.4 notes, GPT-6.1 Sol first
- 1332e9c Merge pull request #9226 from code-yeongyu/fix/9223-lsp-install-nudge-scope
- 32efbda docs(changes): note the Codex post-edit cache shape for #9223
- 546e1d7 fix(codex-lsp): build the post-edit session cache with the not-installed server set (#9223)
- bbc06b2 Merge remote-tracking branch 'origin/dev' into fix/9223-lsp-install-nudge-scope
- 09ba730 fix(lsp): skip post-edit install nudges outside a project and nudge once per server (#9223)
- d8b85db Merge pull request #9224 from code-yeongyu/feat/9214-gpt-6-1-sol-routing
- d825f9f docs: deep-low and Geeky · Normal run GPT-6.1 Sol at medium (#9214)
- 6bbc112 feat(categories): lead deep-low with gpt-6.1-sol medium (#9214)
- 0be337f feat(model-core): know GPT-6.1 Sol capabilities (#9214)
- ad50233 Merge pull request #9205 from code-yeongyu/fix/desktop-34-quick-zai-xiaomi-app-surface
- e56a538 docs(changes): keep the blank line after dev's entries in the #9202 union merge
bun add -g omo-ai