levelcode/extensions/levelcode-ai/inlineComplete.js at develop · levelcodeai/levelcode · GitHub

History

155 lines (136 loc) · 6.44 KB

Raw

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

/*---------------------------------------------------------------------------------------------

* LevelCode — AI (M2, feature 2: inline tab-completion)

* Ghost-text completions as you type, accepted with Tab. Uses VS Code's stable

* InlineCompletionItemProvider API (no core patches). Requests go directly to the

* provider (Claude / Ollama) — debounced, and cancelled the moment you keep typing.

*--------------------------------------------------------------------------------------------*/

// @ts-check

'use strict';

const vscode = require('vscode');

const PREFIX_MAX = 2500; // chars of context before the cursor (smaller = faster first token)

const SUFFIX_MAX = 1000; // chars of context after the cursor

const MAX_TOKENS = 160; // keep completions short & snappy

const SYSTEM_PROMPT =

'You are a code completion engine inside the LevelCode editor. You are given the code ' +

'before the cursor and the code after it, with the insertion point marked <CURSOR>. ' +

'Reply with ONLY the raw code that should be inserted at <CURSOR> to continue the code ' +

'naturally. Do NOT repeat code that already exists before or after the cursor. Do NOT ' +

'wrap your answer in markdown fences. Do NOT add explanations. If no completion is ' +

'appropriate, reply with an empty string.';

/** @param {vscode.CancellationToken} token */

function sleep(ms, token) {

return new Promise((resolve) => {

const t = setTimeout(resolve, ms);

token.onCancellationRequested(() => { clearTimeout(t); resolve(undefined); });

});

}

/** Build prefix/suffix context windows around the cursor. */

function buildContext(document, position) {

const start = new vscode.Position(0, 0);

const lastLine = document.lineAt(document.lineCount - 1);

const end = lastLine.range.end;

const pre = document.getText(new vscode.Range(start, position));

const suf = document.getText(new vscode.Range(position, end));

return { prefix: pre.slice(-PREFIX_MAX), suffix: suf.slice(0, SUFFIX_MAX) };

}

/** Strip accidental markdown fences / leading language tags the model may add. */

function clean(text) {

if (!text) { return ''; }

let t = text;

const fence = String.fromCharCode(96, 96, 96);

if (t.indexOf(fence) >= 0) {

// Take the content of the first fenced block if the whole thing got wrapped.

const m = t.match(new RegExp(fence + '[a-zA-Z0-9_+-]*\\n?([\\s\\S]*?)' + fence));

if (m) { t = m[1]; }

else { t = t.split(fence).join(''); }

}

return t;

}

/**

* @param {vscode.ExtensionContext} context

* @param {{ aiConfig: () => vscode.WorkspaceConfiguration,

* prepProviderRequest: (o?:any) => Promise<any>,

* complete: Function, fastCompletionModel: (providerId:string)=>(string|null) }} deps

function registerInlineComplete(context, deps) {

const { aiConfig, prepProviderRequest, complete, fastCompletionModel } = deps;

// --- status bar toggle ---------------------------------------------------

const status = vscode.window.createStatusBarItem(vscode.StatusBarAlignment.Right, 50);

status.command = 'levelcode.ai.toggleCompletions';

function refreshStatus() {

const on = aiConfig().get('completions.enabled', true);

status.text = on ? '$(sparkle) AI' : '$(circle-slash) AI';

status.tooltip = on

? 'LevelCode AI inline completions: ON (click to disable)'

: 'LevelCode AI inline completions: OFF (click to enable)';

status.show();

}

refreshStatus();

context.subscriptions.push(

status,

vscode.commands.registerCommand('levelcode.ai.toggleCompletions', async () => {

const cfg = aiConfig();

const next = !cfg.get('completions.enabled', true);

await cfg.update('completions.enabled', next, vscode.ConfigurationTarget.Global);

refreshStatus();

vscode.window.setStatusBarMessage('LevelCode AI completions ' + (next ? 'enabled' : 'disabled'), 2000);

}),

vscode.workspace.onDidChangeConfiguration((e) => {

if (e.affectsConfiguration('levelcode.ai.completions')) { refreshStatus(); }

})

);

// --- the provider --------------------------------------------------------

/** @type {vscode.InlineCompletionItemProvider} */

const provider = {

async provideInlineCompletionItems(document, position, ctx, token) {

const cfg = aiConfig();

if (!cfg.get('completions.enabled', true)) { return null; }

if (document.uri.scheme !== 'file' && document.uri.scheme !== 'untitled') { return null; }

// Debounce: if the user keeps typing, this token is cancelled and we bail.

const debounce = Math.max(0, cfg.get('completions.debounce', 150));

await sleep(debounce, token);

if (token.isCancellationRequested) { return null; }

const { prefix, suffix } = buildContext(document, position);

if (!prefix.trim() && !suffix.trim()) { return null; }

const userContent =

'Language: ' + (document.languageId || 'plaintext') + '\n' +

'Insert code at <CURSOR>. Reply with ONLY the text to insert.\n\n' +

prefix + '<CURSOR>' + suffix;

const ac = new AbortController();

const sub = token.onCancellationRequested(() => ac.abort());

let text = '';

try {

// Silent (prompt:false): never pop a key dialog while the user is typing.

const req = await prepProviderRequest({ prompt: false });

if (!req.ok || token.isCancellationRequested) { return null; }

// Claude uses the configurable completions model; other providers use their curated fast

// model (gpt-4o-mini, llama-3.1-8b-instant, …) when one exists, else the active chat model.

const model = req.providerId === 'claude'

? cfg.get('completions.model', 'claude-haiku-4-5-20251001')

: (fastCompletionModel(req.providerId) || req.model);

text = await complete({

providerId: req.providerId, apiKey: req.apiKey, baseURL: req.baseURL,

model, maxTokens: MAX_TOKENS, system: SYSTEM_PROMPT,

messages: [{ role: 'user', content: userContent }], signal: ac.signal

});

} catch (e) {

// Network/abort/key errors are silent — inline completion must never nag.

return null;

} finally {

sub.dispose();

}

if (token.isCancellationRequested) { return null; }

const insert = clean(text);

if (!insert.trim()) { return null; }

// Avoid suggesting text that just duplicates what's already after the cursor.

if (suffix.startsWith(insert)) { return null; }

const item = new vscode.InlineCompletionItem(insert, new vscode.Range(position, position));

return [item];

}

};

context.subscriptions.push(

vscode.languages.registerInlineCompletionItemProvider({ pattern: '**' }, provider)

);

}

module.exports = { registerInlineComplete };

levelcode/extensions/levelcode-ai/inlineComplete.js at develop · levelcodeai/levelcode · GitHub

Uh oh!

Expand file tree

Breadcrumbs

inlineComplete.js

Latest commit

History

Breadcrumbs

inlineComplete.js

File metadata and controls