helix-gpt

Code assistant language server for Helix with support for Copilot/OpenAI/Codeium.

Completion example

Code actions example (space + a)

Available code actions: resolveDiagnostics generateDocs improveCode refactorFromComment writeTest

How?

When a trigger character is pressed it will request a completion and use the entire file as context. Default triggers characters: ["{", "(", " "] can be overwritten with --triggerCharacters "{||(|| "

Use ctrl+x to manually trigger completions, and space+a to trigger code actions that only use the selected code as context.

Install

This was made to run with Bun, but you can also use a precompiled binary.

Without Bun

wget https://github.com/leona/helix-gpt/releases/download/0.34/helix-gpt-0.34-x86_64-linux.tar.gz \
-O /tmp/helix-gpt.tar.gz \
&& tar -zxvf /tmp/helix-gpt.tar.gz \
&& mv helix-gpt-0.34-x86_64-linux /usr/bin/helix-gpt \
&& chmod +x /usr/bin/helix-gpt

With Bun (tested with 1.0.25)

wget https://github.com/leona/helix-gpt/releases/download/0.34/helix-gpt-0.34.js -O /usr/bin/helix-gpt

Configuration

You can configure helix-gpt by exposing either the environment variables below, or by passing the command line options directly to helix-gpt in the Helix configuration step.

All configuration options

NOTE: Copilot is the best choice due to the model and implementation.

Environment Variables

OPENAI_API_KEY=123 # Required if using openai handler
COPILOT_API_KEY=123 # Required if using copilot handler
CODEIUM_API_KEY=123 # Not required, will use public API key otherwise.
HANDLER=openai # openai/copilot/codeium

Command Line Arguments

(Add to command = "helix-gpt" in Helix configuration)

--handler openai --openaiKey 123

You can also use:

helix-gpt --authCopilot

To fetch your Copilot token.

Helix Configuration

Example for TypeScript .helix/languages.toml tested with Helix 23.10 (older versions may not support multiple LSPs)

[language-server.gpt]
command = "helix-gpt"

[language-server.ts]
command = "typescript-language-server"
args = ["--stdio"]
language-id = "javascript"

[[language]]
name = "typescript"
language-servers = [
    "ts",
    "gpt"
]

In case you opt out of the precompiled binary, modify as follows:

[language-server.gpt]
command = "bun"
args = ["run", "/app/helix-gpt.js"]

All Done

If there are any issues, refer to the helix-gpt and Helix log files:

tail -f /root/.cache/helix/helix.log
tail -f /app/helix-gpt.log # Or wherever you set --logFile to

Special Thanks

rsc1975 for their Bun Dockerfile.

Todo

Copilot support
Resolve diagnostics code action
Self-hosted model support (partial support if they are openai compliant)
Inline completion provider (pending support from Helix)
Single config for all languages (pending #9318)
Support workspace commands to toggle functionality (pending Helix support for merging workspace commands)
Increase test coverage
Async load completions to show other language server results immediately (pending Helix support)
Improve recovery from errors as it can leave the editor unusable sometimes