The problems with today's agentic code editors that CodeMeld sets out to fix
When an editor works with only one company’s models, your work stops whenever that provider has a problem or decides to lock you out, as one company found when it lost access to Claude overnight.
CodeMeld works with OpenRouter, Venice or your ChatGPT plan. If one stops working for you, switch to another and keep going.
Most agentic editors assume git, and keep reaching for git even when you have told them you use something else.
CodeMeld works with Git, Mercurial and Jujutsu out of the box, and you can add any other version control system yourself.
You approve two commands, and then the agent runs them together or writes one slightly differently, and you are asked all over again.
In CodeMeld you approve a command once in a folder, and you are not asked again when the same command is written differently.
“Don’t ask again” rarely says for how long, or where. An approval may vanish when the session ends, or quietly cover folders you never meant it to.
CodeMeld tells you how long each approval lasts and lets you choose. An approval applies only to the folder where you gave it and the folders inside it.
When prompts become too many, the usual answer is a mode that approves almost anything, by rules you cannot see.
CodeMeld has no approve-everything mode. You get fewer prompts because approvals are remembered and apply where they should, not because checks are switched off.
CodeMeld is open source: github.com/commonforge-dev/codemeld