RU

fish_harness: AI coding agent with CLI, IDE and desktop on one engine

Published: 2026-10-18 · Author: AI Release · @ai_release1
fish_harness: AI coding agent with CLI, IDE and desktop on one engine

⚡ The gist in 5 seconds - The zelenka84/fish_harness project has been published on GitHub — an AI agent for programming with three interfaces on top of a single engine: a desktop app, a Cursor-style IDE, and a terminal CLI. - It supports Claude, OpenAI, OpenRouter, Gemini, local models, and any compatible API. - The source code is open on GitHub; the version, license, and system requirements are not specified in the announcement. ### 🔍 What was found The zelenka84/fish_harness repository describes itself as an "AI coding agent with three faces over one engine" — that is, a single agent engine to which three different frontends connect: a graphical desktop app, a Cursor-style editor, and a console CLI. The author is publishing the project openly on GitHub; the link leads to the repository page. The key feature, according to the description, is multi-provider support. The agent is claimed to work with Claude, OpenAI, OpenRouter, Gemini, local models, "or any compatible" backend — meaning there is no lock-in to a single vendor: the provider and model are switched at the level of the same shared engine, not in a separate client. Configuration details, the list of supported model versions, and the method for connecting local weights are not disclosed in the description — you will have to check the repository itself for those. ### 💡 Why it matters The coding agent niche is already crowded: Cursor, Claude Code, Gemini CLI, Aider, and dozens of API wrappers. fish_harness's bet is not on an exclusive model, but on the "one engine — three interfaces" architecture, with the ability to move the agent to local weights or your own API-compatible endpoint. For those who do not want to be tied to a specific vendor's subscription and prefer to keep some tasks on their own models, this could be a viable option — but the claimed capabilities are so far backed only by the repository description, without benchmarks or comparisons.

🤖 AI summary
#AI-агент#open#CLI#LLM#разработка#AI#нейросети#ChatGPT
← PreviousOpenAI is developing humanoid robots: betting on the AI brain with a 2027 timelineNext →Gemini 4 Argon is Cheaper Than Opus... But the Catch?

Source: github.com · post in Telegram