Add initial implementation of noThinkProxy with Docker support and README documentation
This commit is contained in:
213
app.js
213
app.js
@@ -1,20 +1,138 @@
|
||||
// proxy.js - KOMPLETT, mit Tool-Call-Deduplizierung
|
||||
|
||||
const express = require('express');
|
||||
const app = express();
|
||||
|
||||
const OLLAMA_URL = process.env.OLLAMA_URL || 'https://ollama.aquantico.de/api/chat';
|
||||
const OLLAMA_MODEL = process.env.OLLAMA_MODEL || 'qwen3.6:35b-a3b-q4_K_M';
|
||||
const OLLAMA_AUTH = process.env.OLLAMA_AUTH || '324GF44-50AA-4B57-9386-K435DLJ764DFR';
|
||||
const PORT = parseInt(process.env.PORT || '11435', 10);
|
||||
|
||||
const colors = {
|
||||
reset: '\x1b[0m',
|
||||
cyan: '\x1b[36m',
|
||||
green: '\x1b[32m',
|
||||
reset: '\x1b[0m',
|
||||
cyan: '\x1b[36m',
|
||||
green: '\x1b[32m',
|
||||
magenta: '\x1b[35m',
|
||||
yellow: '\x1b[33m',
|
||||
blue: '\x1b[34m',
|
||||
red: '\x1b[31m'
|
||||
yellow: '\x1b[33m',
|
||||
blue: '\x1b[34m',
|
||||
red: '\x1b[31m'
|
||||
};
|
||||
|
||||
app.set('trust proxy', 1);
|
||||
app.use(express.json({ limit: '50mb' }));
|
||||
|
||||
// ── Info-Seite ────────────────────────────────────────────────────────────────
|
||||
|
||||
app.get('/', (req, res) => {
|
||||
const host = `${req.protocol}://${req.get('host')}`;
|
||||
res.setHeader('Content-Type', 'text/html; charset=utf-8');
|
||||
res.send(`<!DOCTYPE html>
|
||||
<html lang="de">
|
||||
<head>
|
||||
<meta charset="UTF-8">
|
||||
<meta name="viewport" content="width=device-width, initial-scale=1">
|
||||
<title>noThinkProxy</title>
|
||||
<style>
|
||||
*{box-sizing:border-box;margin:0;padding:0}
|
||||
body{font-family:system-ui,sans-serif;background:#0f0f0f;color:#e0e0e0;padding:2rem}
|
||||
h1{color:#a78bfa;font-size:1.8rem;margin-bottom:.4rem}
|
||||
h2{color:#7dd3fc;font-size:1.1rem;margin:2rem 0 .6rem}
|
||||
p{color:#9ca3af;line-height:1.6;margin-bottom:.8rem}
|
||||
code{background:#1e1e2e;color:#cba6f7;padding:.15rem .4rem;border-radius:.25rem;font-size:.9rem}
|
||||
pre{background:#1e1e2e;border:1px solid #333;border-radius:.5rem;padding:1rem;overflow-x:auto;margin:.5rem 0 1rem}
|
||||
pre code{background:none;padding:0;color:#a6e3a1}
|
||||
.badge{display:inline-block;background:#1e3a5f;color:#7dd3fc;border-radius:.25rem;padding:.1rem .5rem;font-size:.8rem;margin-left:.5rem}
|
||||
table{width:100%;border-collapse:collapse;margin:.5rem 0 1rem}
|
||||
td,th{border:1px solid #333;padding:.4rem .8rem;text-align:left}
|
||||
th{background:#1e1e2e;color:#7dd3fc}
|
||||
.copy-btn{cursor:pointer;background:#2d2d4e;border:1px solid #555;color:#cba6f7;padding:.2rem .6rem;border-radius:.25rem;font-size:.75rem;margin-left:.5rem}
|
||||
</style>
|
||||
</head>
|
||||
<body>
|
||||
<h1>noThinkProxy <span class="badge">v1.0</span></h1>
|
||||
<p>Anthropic-API → Ollama-Proxy · Think-Modus deaktiviert · Modell-Substitution aktiv</p>
|
||||
|
||||
<h2>Aktuelle Konfiguration</h2>
|
||||
<table>
|
||||
<tr><th>Parameter</th><th>Wert</th></tr>
|
||||
<tr><td>Ollama URL</td><td><code>${OLLAMA_URL.replace(/\/api\/chat$/, '')}</code></td></tr>
|
||||
<tr><td>Modell</td><td><code>${OLLAMA_MODEL}</code></td></tr>
|
||||
<tr><td>Kontext</td><td><code>262144 Token (256k)</code></td></tr>
|
||||
<tr><td>Think</td><td><code>false</code></td></tr>
|
||||
<tr><td>Proxy-URL</td><td><code>${host}</code></td></tr>
|
||||
</table>
|
||||
|
||||
<h2>localclaude installieren</h2>
|
||||
<p>Installiert das Script <code>localclaude</code> nach <code>/usr/local/bin</code> (oder <code>~/.local/bin</code>):</p>
|
||||
<pre><code>curl -fsSL ${host}/install.sh | bash</code></pre>
|
||||
|
||||
<h2>Starten</h2>
|
||||
<pre><code>localclaude</code></pre>
|
||||
<p><code>localclaude</code> setzt automatisch <code>ANTHROPIC_BASE_URL=${host}</code> und ruft <code>claude</code> auf.</p>
|
||||
|
||||
<h2>API-Endpunkt</h2>
|
||||
<pre><code>POST ${host}/v1/messages</code></pre>
|
||||
<p>Kompatibel mit dem Anthropic SDK. Alle <code>claude-*</code> Modellnamen werden automatisch auf <code>${OLLAMA_MODEL}</code> umgeleitet.</p>
|
||||
</body>
|
||||
</html>`);
|
||||
});
|
||||
|
||||
// ── Install-Script ────────────────────────────────────────────────────────────
|
||||
|
||||
app.get('/install.sh', (req, res) => {
|
||||
const host = `${req.protocol}://${req.get('host')}`;
|
||||
res.setHeader('Content-Type', 'text/plain; charset=utf-8');
|
||||
res.send(`#!/usr/bin/env bash
|
||||
set -euo pipefail
|
||||
|
||||
PROXY_URL="${host}"
|
||||
INSTALL_DIR="/usr/local/bin"
|
||||
NEEDS_PATH_UPDATE=false
|
||||
|
||||
echo ""
|
||||
echo "=== noThinkProxy · localclaude Installer ==="
|
||||
echo ""
|
||||
|
||||
# Zielverzeichnis bestimmen (ohne sudo → ~/.local/bin)
|
||||
if [ ! -w "\$INSTALL_DIR" ]; then
|
||||
INSTALL_DIR="\$HOME/.local/bin"
|
||||
mkdir -p "\$INSTALL_DIR"
|
||||
fi
|
||||
|
||||
# localclaude-Script schreiben
|
||||
cat > "\$INSTALL_DIR/localclaude" <<'SCRIPT'
|
||||
#!/usr/bin/env bash
|
||||
export ANTHROPIC_BASE_URL="${host}"
|
||||
exec claude "\$@"
|
||||
SCRIPT
|
||||
|
||||
chmod +x "\$INSTALL_DIR/localclaude"
|
||||
|
||||
# PATH prüfen und ggf. in Shell-Config eintragen
|
||||
if ! echo "\$PATH" | grep -q "\$INSTALL_DIR"; then
|
||||
NEEDS_PATH_UPDATE=true
|
||||
echo "» Trage \$INSTALL_DIR in ~/.bashrc und ~/.zshrc ein..."
|
||||
echo "export PATH=\\"\$INSTALL_DIR:\$PATH\\"" >> "\$HOME/.bashrc"
|
||||
echo "export PATH=\\"\$INSTALL_DIR:\$PATH\\"" >> "\$HOME/.zshrc" 2>/dev/null || true
|
||||
fi
|
||||
|
||||
echo "✓ localclaude installiert in \$INSTALL_DIR"
|
||||
echo ""
|
||||
if [ "\$NEEDS_PATH_UPDATE" = "true" ]; then
|
||||
echo "────────────────────────────────────────────"
|
||||
echo "Führe diesen Befehl jetzt aus damit localclaude sofort verfügbar ist:"
|
||||
echo ""
|
||||
echo " export PATH=\\"\$INSTALL_DIR:\$PATH\\""
|
||||
echo ""
|
||||
echo "In neuen Shell-Sessions ist es automatisch verfügbar."
|
||||
echo "────────────────────────────────────────────"
|
||||
else
|
||||
echo "Starte mit: localclaude"
|
||||
fi
|
||||
echo ""
|
||||
`);
|
||||
});
|
||||
|
||||
// ── Hilfsfunktionen ───────────────────────────────────────────────────────────
|
||||
|
||||
function sanitizeToolSchema(schema) {
|
||||
if (!schema || typeof schema !== 'object') {
|
||||
return { type: 'object', properties: {} };
|
||||
@@ -70,7 +188,6 @@ function stringifyToolResultContent(content) {
|
||||
return JSON.stringify(content);
|
||||
}
|
||||
|
||||
// BUG 3 FIX: tool_use → Ollama tool_calls, tool_result → role:tool
|
||||
function convertAnthropicToOllama(anthropicBody) {
|
||||
const ollamaMessages = [];
|
||||
|
||||
@@ -116,7 +233,6 @@ function convertAnthropicToOllama(anthropicBody) {
|
||||
ollamaMessages.push(assistantMsg);
|
||||
|
||||
} else {
|
||||
// user messages: text bleibt als user, tool_result wird zu role:tool
|
||||
const pendingText = [];
|
||||
|
||||
for (const item of msg.content) {
|
||||
@@ -178,9 +294,7 @@ function parseToolArguments(args) {
|
||||
}
|
||||
}
|
||||
|
||||
if (typeof args === 'object') {
|
||||
return args;
|
||||
}
|
||||
if (typeof args === 'object') return args;
|
||||
|
||||
return {};
|
||||
}
|
||||
@@ -193,6 +307,8 @@ function makeToolDedupeKey(tc) {
|
||||
return `${name}:${argsString}`;
|
||||
}
|
||||
|
||||
// ── Response-Handler ──────────────────────────────────────────────────────────
|
||||
|
||||
async function handleResponse(response, anthropicBody, res, requestNum) {
|
||||
res.setHeader('Content-Type', 'text/event-stream');
|
||||
res.setHeader('Cache-Control', 'no-cache');
|
||||
@@ -200,7 +316,6 @@ async function handleResponse(response, anthropicBody, res, requestNum) {
|
||||
|
||||
const messageId = 'msg_' + requestNum;
|
||||
|
||||
// BUG 2 FIX: usage.input_tokens und stop_reason hinzugefügt
|
||||
res.write(`event: message_start\ndata: ${JSON.stringify({
|
||||
type: 'message_start',
|
||||
message: {
|
||||
@@ -226,7 +341,6 @@ async function handleResponse(response, anthropicBody, res, requestNum) {
|
||||
let messageFinished = false;
|
||||
let buffer = '';
|
||||
|
||||
// BUG 1 FIX: Chunk-Verarbeitung als Funktion, damit final buffer ebenfalls verarbeitet wird
|
||||
function processChunk(data) {
|
||||
if (messageFinished) return;
|
||||
|
||||
@@ -253,21 +367,13 @@ async function handleResponse(response, anthropicBody, res, requestNum) {
|
||||
res.write(`event: content_block_start\ndata: ${JSON.stringify({
|
||||
type: 'content_block_start',
|
||||
index: currentBlockIndex,
|
||||
content_block: {
|
||||
type: 'tool_use',
|
||||
id: toolUseId,
|
||||
name: toolName,
|
||||
input: {}
|
||||
}
|
||||
content_block: { type: 'tool_use', id: toolUseId, name: toolName, input: {} }
|
||||
})}\n\n`);
|
||||
|
||||
res.write(`event: content_block_delta\ndata: ${JSON.stringify({
|
||||
type: 'content_block_delta',
|
||||
index: currentBlockIndex,
|
||||
delta: {
|
||||
type: 'input_json_delta',
|
||||
partial_json: JSON.stringify(toolInput)
|
||||
}
|
||||
delta: { type: 'input_json_delta', partial_json: JSON.stringify(toolInput) }
|
||||
})}\n\n`);
|
||||
|
||||
res.write(`event: content_block_stop\ndata: ${JSON.stringify({
|
||||
@@ -286,10 +392,7 @@ async function handleResponse(response, anthropicBody, res, requestNum) {
|
||||
res.write(`event: content_block_start\ndata: ${JSON.stringify({
|
||||
type: 'content_block_start',
|
||||
index: currentBlockIndex,
|
||||
content_block: {
|
||||
type: 'text',
|
||||
text: ''
|
||||
}
|
||||
content_block: { type: 'text', text: '' }
|
||||
})}\n\n`);
|
||||
|
||||
contentBlocks[currentBlockIndex] = '';
|
||||
@@ -300,10 +403,7 @@ async function handleResponse(response, anthropicBody, res, requestNum) {
|
||||
res.write(`event: content_block_delta\ndata: ${JSON.stringify({
|
||||
type: 'content_block_delta',
|
||||
index: currentBlockIndex,
|
||||
delta: {
|
||||
type: 'text_delta',
|
||||
text
|
||||
}
|
||||
delta: { type: 'text_delta', text }
|
||||
})}\n\n`);
|
||||
|
||||
contentBlocks[currentBlockIndex] += text;
|
||||
@@ -321,17 +421,11 @@ async function handleResponse(response, anthropicBody, res, requestNum) {
|
||||
|
||||
res.write(`event: message_delta\ndata: ${JSON.stringify({
|
||||
type: 'message_delta',
|
||||
delta: {
|
||||
stop_reason: emittedToolUse ? 'tool_use' : 'end_turn'
|
||||
},
|
||||
usage: {
|
||||
output_tokens: data.eval_count || 0
|
||||
}
|
||||
delta: { stop_reason: emittedToolUse ? 'tool_use' : 'end_turn' },
|
||||
usage: { output_tokens: data.eval_count || 0 }
|
||||
})}\n\n`);
|
||||
|
||||
res.write(`event: message_stop\ndata: ${JSON.stringify({
|
||||
type: 'message_stop'
|
||||
})}\n\n`);
|
||||
res.write(`event: message_stop\ndata: ${JSON.stringify({ type: 'message_stop' })}\n\n`);
|
||||
|
||||
console.log(`${colors.green}✓${colors.reset}\n`);
|
||||
}
|
||||
@@ -359,7 +453,6 @@ async function handleResponse(response, anthropicBody, res, requestNum) {
|
||||
}
|
||||
}
|
||||
|
||||
// BUG 1 FIX: letzten gepufferten Chunk verarbeiten (kein \n am Ende)
|
||||
if (buffer.trim()) {
|
||||
try {
|
||||
processChunk(JSON.parse(buffer.trim()));
|
||||
@@ -379,22 +472,18 @@ async function handleResponse(response, anthropicBody, res, requestNum) {
|
||||
|
||||
res.write(`event: message_delta\ndata: ${JSON.stringify({
|
||||
type: 'message_delta',
|
||||
delta: {
|
||||
stop_reason: emittedToolUse ? 'tool_use' : 'end_turn'
|
||||
},
|
||||
usage: {
|
||||
output_tokens: 0
|
||||
}
|
||||
delta: { stop_reason: emittedToolUse ? 'tool_use' : 'end_turn' },
|
||||
usage: { output_tokens: 0 }
|
||||
})}\n\n`);
|
||||
|
||||
res.write(`event: message_stop\ndata: ${JSON.stringify({
|
||||
type: 'message_stop'
|
||||
})}\n\n`);
|
||||
res.write(`event: message_stop\ndata: ${JSON.stringify({ type: 'message_stop' })}\n\n`);
|
||||
}
|
||||
|
||||
res.end();
|
||||
}
|
||||
|
||||
// ── Proxy-Endpunkt ────────────────────────────────────────────────────────────
|
||||
|
||||
app.post('/v1/messages', async (req, res) => {
|
||||
const requestNum = Date.now();
|
||||
|
||||
@@ -404,20 +493,20 @@ app.post('/v1/messages', async (req, res) => {
|
||||
const anthropicBody = req.body;
|
||||
|
||||
if (anthropicBody.model?.startsWith('claude-')) {
|
||||
anthropicBody.model = 'qwen3.6:35b-a3b-q4_K_M';
|
||||
anthropicBody.model = OLLAMA_MODEL;
|
||||
}
|
||||
|
||||
const ollamaBody = convertAnthropicToOllama(anthropicBody);
|
||||
|
||||
console.log(
|
||||
`${colors.magenta}[msgs=${ollamaBody.messages.length}, tools=${ollamaBody.tools?.length || 0}, ctx=256k, think=false]${colors.reset}`
|
||||
`${colors.magenta}[msgs=${ollamaBody.messages.length}, tools=${ollamaBody.tools?.length || 0}, ctx=256k, think=false, model=${OLLAMA_MODEL}]${colors.reset}`
|
||||
);
|
||||
|
||||
const response = await fetch('https://ollama.aquantico.de/api/chat', {
|
||||
const response = await fetch(OLLAMA_URL, {
|
||||
method: 'POST',
|
||||
headers: {
|
||||
'Content-Type': 'application/json',
|
||||
'Authorization': 'Bearer 324GF44-50AA-4B57-9386-K435DLJ764DFR'
|
||||
'Authorization': `Bearer ${OLLAMA_AUTH}`
|
||||
},
|
||||
body: JSON.stringify(ollamaBody)
|
||||
});
|
||||
@@ -435,10 +524,7 @@ app.post('/v1/messages', async (req, res) => {
|
||||
if (!res.headersSent) {
|
||||
res.status(500).json({
|
||||
type: 'error',
|
||||
error: {
|
||||
type: 'api_error',
|
||||
message: error.message
|
||||
}
|
||||
error: { type: 'api_error', message: error.message }
|
||||
});
|
||||
} else {
|
||||
res.end();
|
||||
@@ -446,6 +532,9 @@ app.post('/v1/messages', async (req, res) => {
|
||||
}
|
||||
});
|
||||
|
||||
app.listen(11435, () => {
|
||||
console.log(`${colors.magenta}Proxy: localhost:11435 (256k ctx, think=false)${colors.reset}\n`);
|
||||
app.listen(PORT, () => {
|
||||
console.log(`${colors.magenta}noThinkProxy: localhost:${PORT}${colors.reset}`);
|
||||
console.log(`${colors.cyan} Ollama : ${OLLAMA_URL}${colors.reset}`);
|
||||
console.log(`${colors.cyan} Modell : ${OLLAMA_MODEL}${colors.reset}`);
|
||||
console.log(`${colors.cyan} Ctx : 256k Think: false${colors.reset}\n`);
|
||||
});
|
||||
|
||||
Reference in New Issue
Block a user