@paularskii@banarum0 Прямо на скрине написано другое. Сам интерфейс доступен в платформе много лет, но как и с vscode платформа требует написания плагина. Это нужно как раз потому что полезные LSP сервера работают в основном на кастомных расширениях
@paularskii@banarum0 У него конечно же нет интерфейса LSP, но это полноценная поддержка. PSI позволяет поддержать намного более широкий суперсет по сравнению LSP. Более того, никакой фактически клиент не ограничивается стандартом LSP, это все кастомные расширения протокола
@khurramijaz@johncrickett@jetbrains What prevents using Junie in zed or any other acp client? Also, what is wrong with using openai sub with it (it works everywhere without any limitations).
What would even tell pi that it is running with an ACP UI for it to become dumb?
@johnny5__alive@AnvarRamazanov@SanSYS Nice, I also like it. Note that this separate standalone app might not get a lot of updates in the future. Seems like users prefer Air in IntelliJ, so it will likely get more attention in the future. It has the same flexibility as well though
@johnny5__alive@SanSYS This was and still is available since AI Assistant, via custom providers. With ACP you can also configure any ACP-compatible agent and there are many designed specifically for self hosted models. This way you don’t get forced into a specific agent harness, it is more flexible
@bretajohnson@kskrygan@ares_mheinke Afaik, at this time when asked directly it is a hard stop from their side, most likely the updates would be here: https://t.co/jmb5koz990
You can get it to work if you externally manage the auth personally, but this is same grey area
@jvmusin@kskrygan Air would use codex compaction when you run codex and overall whatever the agent of your choice does natively (including but not limited to opencode).
@iterm11_@anastasiak2512@kskrygan Afaik cursor’s acp implementation is quite lacking, and these controls needs to be exposed by cursor agent first, client alone can’t do much
@bretajohnson@kskrygan@ares_mheinke Codex - ofc yes, Claude - no. This is a tos issue, not an engineering problem. With all other agents you can use auth handled by the agent itself natively
@tomislav_uzelac@jetbrains It isn’t. What IntelliJ provides for sure is available with the IDE at no additional cost (MCP). This is a separate offering for semantic index (based on embeddings), IDEs do not have this and local hardware typically doesn’t handle workload fast enough to be useful
@cubdesign Most probably the agent doesn’t do the editing through MCP, so the IDE can’t know that you want the file to re-fresh and re-index. Just do on-demand refresh via an icon in vcs tw. IDE maintains consistent state, contrary to simple editors, you don’t want to do it on every change
@Reineir10@jetbrains Agents aren’t perfect, contrary to what some could say. AI Assistant is a platform, agents themselves are supplied by labs. Claude Agent - Anthropic, Codex - OpenAI. Chat option isn’t even an agent (no loop intentionally). Consider creating a ticket if it seems like an outage
@JusBili @jetbrains@huggingface@kilocode@cline The model is trained from scratch, not from Qwen or any other existing base, base model is also available on HF. Junie at this moment targets the biggest models only. There is no API access for this, since it is meant for local use-cases