README.md


1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50

# Hexai

![HexAI Small Logo](hexai-small.png)

Hexai, the AI LSP for the Helix editor.

At the moment this project is only in the proof of concept phase.

## LLM provider

Hexai exposes a simple LLM provider interface and uses OpenAI by default for
code completion when `OPENAI_API_KEY` is present in the environment.

- Required: set `OPENAI_API_KEY` to your OpenAI API key.
- Optional: set `OPENAI_MODEL`.
- Optional: set `OPENAI_BASE_URL` to point at a compatible endpoint.

If no key is configured, Hexai will fall back to a basic, local completion.

## CLI usage and configuration

- Run LSP server over stdio:
  - `hexai`

- Completion settings:
  - `-max-tokens`: maximum tokens for LLM completions. If the flag isn’t provided, `HEXAI_MAX_TOKENS` is used when set.
  - `-context-mode`: how much additional context to include with completion prompts (If the flag isn’t provided, `HEXAI_CONTEXT_MODE` is used when set). One of:
    - `minimal`: no extra context
    - `window`: include a sliding window around the cursor
    - `file-on-new-func`ude the full file only when defining a new function (cursor before the opening `{`)
    - `always-full`: always include the full file (may be slower/costly)
  - `-context-window-lines`: line count for the sliding window when `context-mode=window`.
  - `-max-context-tokens`: budget for additional context tokens. If the flag isn’t provided, `HEXAI_MAX_CONTEXT_TOKENS` is used when set.

Notes:
- Token estimation for truncation uses a simple 4 chars/token heuristic.
- Full-file context is only included by default when defining a new function to balance quality, latency, and cost.

### Flags quick reference

| Flag                    | Env override               | Description                                        |
|-------------------------|----------------------------|----------------------------------------------------|
| `-stdio`                | —                          | Run as LSP over stdio (only supported mode).       |
| `-log`                  | —                          | Path to log file (optional).                       |
| `-max-tokens`           | `HEXAI_MAX_TOKENS`         | Max tokens for LLM completions.                    |
| `-context-mode`         | `HEXAI_CONTEXT_MODE`       | `minimal` `window` `file-on-new-func` `always-full` |
| `-context-window-lines` | `HEXAI_CONTEXT_WINDOW_LINES` | Lines around cursor when using `window` mode.      |
| `-max-context-tokens`   | `HEXAI_MAX_CONTEXT_TOKENS` | Token budget for additional context.               |
| `-log-preview-limit`    | `HEXAI_LOG_PREVIEW_LIMIT`  | Limit characters shown in LLM preview logs.        |
| `-no-disk-io`           | `HEXAI_NO_DISK_IO`         | Disallow any disk reads for context.               |