avante.nvim

You will notice how little ceremony there is. Suggestions arrive and you apply them with one action, which is the deciding daily trait, because the thing that kills editor agents is the friction between reading a proposal and getting it into the file. Project instruction files mean your conventions survive between sessions, and an optional local index keeps the model pointed at the right part of a large tree.

Pick it if you want that Cursor-shaped experience inside your existing editor. Pick CodeCompanion.nvim when you want conversations and workflows rather than fast inline application.

The structural risk is that the specification lives somewhere else. Emulating another product's behaviour means the reference implementation belongs to a company with no interest in this one, so every interaction change upstream becomes either churn here or a growing gap. The installation also compiles native code through a build step in the plugin manager, which turns a plugin update into a toolchain problem on machines that lack one.

What it does right: project instruction files put the conventions in the repository, where every contributor sees them.

1. Context gathering offers an optional local retrieval index, which keeps embeddings on the machine and makes the cost of that choice explicit rather than hidden in a service. 2. Edit application is a single confirmed action on a proposed change, so the human remains the commit gate. 3. Command execution is not implemented here at all; it arrives through the Agent Client Protocol, which hosts an external agent's loop inside the editor.

No benchmark is published. The observation: delegating the agent loop rather than writing one is the reason this plugin will outlive several of its competitors.

The distribution is remarkable and entirely unmonetised. An audience that size is exactly what a seed-stage company spends two years buying, and here it sits behind a permissive licence with no hosted tier, no paid edition and nothing that resembles a funnel. That is either principled or unfinished, and from the outside the two look identical.

Likely path: a hosted or supported edition, or absorption of the maintainers by a company that wants the audience. Position: adopt the plugin, and watch for the day a signup link appears in the README, because that is when the terms change.

The demo is fast and the audience is narrow. Cost at sixty seats is zero for the plugin and whatever inference each engineer's key consumes, which is unbudgeted and unmeasured because there is no console. Identity, retention and telemetry are not applicable, which sounds like a win until security asks who can answer a questionnaire and the answer is nobody. Nothing here runs unattended, so it contributes nothing to our pipelines.

Onboarding is trivial for the four people already using this editor. Not yet, as a standard.

Fully local is the headline. Both a local server and a direct inference runtime are listed providers, so this works air-gapped with nothing leaving the machine, and when I do want a hosted model my key goes in my own config. The licence is permissive, so a fork is legal and stays mine.

What is missing is the tool protocol. There is no client here, so my existing servers do not attach and I have to reach them through whichever external agent I connect instead. That is a detour, and given the source is readable it is a detour I could remove.