Hemmingway-1 has an OpenAI-compatible API. So it should work in any tool that lets you paste a base URL.

Mostly it does. But two of the tools you would try first can’t use it at all, and one of the others supports it only on a forum’s word.

I went through the documentation of each.

Here is the short version. By their own docs, Copilot for Obsidian, VS Code’s built-in custom endpoint, Continue and Cline can all talk to Hemmingway-1. Cursor has a base-URL override that its official docs do not mention. Codex CLI and Claude Code cannot use it.

One note before the list. I have not tested any of these connections. Every setting below comes from the tool’s own documentation, read on September 21 and 22, 2026.

Disclosure: I build StashBase, which offers Hemmingway as a built-in writing skill. The last section says when that is the simpler route.

What do you need from Hemmingway?

Three values and a key.

(1) The three values.

Base URL   https://hemmingway.io/v1
Model      hemmingway-27b
API key    hemmingway_live_...

They come from the API docs. Some tools want the full endpoint instead: https://hemmingway.io/v1/chat/completions.

(2) A key costs $5 to start.

Sign in at the API platform, create a key, and add credit from $5. The price is $0.24 per million input tokens and $0.90 per million output tokens. I found no free API credit in the docs.

If you pay for the Hemmingway app on Plus or above, you can make a key that draws on your plan instead.

(3) The key is yours alone.

The terms say the service is “priced for one person’s own work.” Do not put your key in a plugin you share, or in a web page.

(4) API requests are not kept.

The privacy policy: “Requests made with an API key are not kept.”

Does it work in Obsidian?

Yes, with Copilot for Obsidian.

(5) Copilot for Obsidian has a custom provider.

It is documented under BYOK, and needs no Copilot licence.

  1. Open Settings → Copilot → BYOK.
  2. Click Add a provider, then Add a custom provider.
  3. Enter the Base URL and the API key.
  4. Under Model ID, enter hemmingway-27b and click Add.
  5. Click Test, then Save.

(6) If Test passes and chat fails, turn on CORS.

The docs cover this case: edit the provider and turn on Enable CORS. Replies then arrive whole instead of streaming.

(7) Smart Composer documents it too.

Its wiki: Settings → Smart Composer → Providers → Add custom provider, with the type OpenAI Compatible. Its last release was in January 2026, so it is quieter than Copilot.

(8) Two well-known plugins I would skip.

Smart Connections moved its API chat into a paid plugin, at $30 a month. Text Generator has a custom provider in its code and no page that documents it.

Does it work in VS Code?

Yes, three ways.

(9) VS Code has it built in.

You need no extension and no Copilot plan. The docs: run Chat: Manage Language Models, select Add Models, then Custom Endpoint, and choose the Chat Completions type. VS Code opens chatLanguageModels.json:

[
  {
    "name": "Hemmingway",
    "vendor": "customendpoint",
    "apiKey": "${input:hemmingwayApiKey}",
    "apiType": "chat-completions",
    "models": [
      {
        "id": "hemmingway-27b",
        "name": "Hemmingway-1",
        "url": "https://hemmingway.io/v1/chat/completions",
        "toolCalling": true,
        "contextWindow": 262144,
        "maxOutputTokens": 32768
      }
    ]
  }
]

The context figure is the one Hemmingway advertises for the API. Its docs state no separate input limit.

This covers chat. Inline code suggestions still need a GitHub account.

(10) Continue takes four lines.

From Continue’s docs, in config.yaml:

models:
  - name: Hemmingway-1
    provider: openai
    model: hemmingway-27b
    apiBase: https://hemmingway.io/v1
    apiKey: <YOUR_API_KEY>

(11) Cline takes four fields.

Cline’s docs: set API Provider to OpenAI Compatible, then fill in Base URL, API Key and Model. Set Max Output Tokens to 32768.

Roo Code had the same settings. Its repository was archived in May 2026, so I would not start with it.

Does it work in Cursor?

Maybe. The setting exists, and Cursor’s official docs do not mention it.

(12) The override is confirmed only on the forum.

Cursor’s API keys page lists OpenAI, Anthropic, Google, Azure and Bedrock. It says nothing about a custom base URL.

A Cursor staff member wrote on the forum in September 2026: “BYOK via Override OpenAI Base URL is an officially supported setup, and we’re not removing it.”

(13) It comes with three catches.

While the override is on, every request goes to your endpoint, including the models from your Cursor subscription. Staff suggest turning it off to get those back.

Cursor sometimes sends a request format that Hemmingway’s API does not document. Staff say this “can break” endpoints like it.

And your key passes through Cursor’s servers on every request. The official page says so.

For writing, I would not fight this. Use VS Code’s built-in route.

What about Codex and Claude Code?

No, for both.

(14) Codex CLI speaks a protocol Hemmingway does not offer.

Codex lets you add a custom provider. But its config reference says of the protocol: “responses is the only supported value.” Hemmingway’s docs list two endpoints, chat completions and the model list.

(15) Claude Code does not support other models.

Anthropic’s docs: it “doesn’t support routing Claude Code to non-Claude models through any gateway.”

What can go wrong?

(16) Thinking eats the reply.

Hemmingway-1 thinks before it answers, at its highest level by default. Its docs say thinking “counts toward max_tokens.”

So a tool that asks for a short reply may get the thinking and no text. This is my reading of the docs, not something I have seen happen. If replies come back empty, raise the tool’s maximum output.

VS Code’s custom endpoint documents a reasoning-effort setting. For the other tools I found no documented way to send reasoning_effort.

Is there a route with no setup?

Two.

The browser demo is free and needs no key.

And in StashBase, Hemmingway is a writing skill rather than a separate agent: ask Claude Code, Codex or the built-in Default Agent for a draft and it writes with Hemmingway-1. You sign in to StashBase, with no Hemmingway key and no base URL, and it drafts beside your notes and sources. See how to choose an agent.

To keep everything on your own machine, run the model locally. llama.cpp serves the same OpenAI-style API at localhost, so the settings above work with the local address in place of Hemmingway’s.