satgate 1.5.1

package/dist/src/proxy/handler.js

@@ -0,0 +1,165 @@
+import { TokenCounter } from './token-counter.js';
+import { createStreamingProxy } from './streaming.js';
+import { resolveModelPrice, tokenCostToSats } from './pricing.js';
+/** Allowed upstream path prefixes — anything else is rejected. */
+const ALLOWED_PATH_PREFIXES = ['/v1/chat/completions', '/v1/completions', '/v1/embeddings'];
+/**
+ * Extracts the model name from an OpenAI-compatible request body.
+ */
+function extractModel(body) {
+    return typeof body.model === 'string' ? body.model : '';
+}
+/**
+ * Creates the AI proxy handler.
+ *
+ * @returns A function that proxies a single inference request to the upstream.
+ */
+export function createProxyHandler(deps) {
+    return async function handleProxy(req, paymentHash) {
+        // Capacity check (before any payment deduction)
+        if (!deps.capacity.tryAcquire()) {
+            return new Response(JSON.stringify({ error: 'Service at capacity, try again later' }), { status: 503, headers: { 'Content-Type': 'application/json', 'Retry-After': '5' } });
+        }
+        const start = Date.now();
+        let streamingResponse = false;
+        try {
+            // Validate request path — only allow known OpenAI-compatible endpoints
+            const requestPath = new URL(req.url).pathname;
+            if (!ALLOWED_PATH_PREFIXES.some(p => requestPath === p)) {
+                return new Response(JSON.stringify({ error: 'Not found' }), { status: 404, headers: { 'Content-Type': 'application/json' } });
+            }
+            // Validate Content-Type
+            const contentType = req.headers.get('content-type');
+            if (!contentType || !contentType.includes('application/json')) {
+                return new Response(JSON.stringify({ error: 'Content-Type must be application/json' }), { status: 415, headers: { 'Content-Type': 'application/json' } });
+            }
+            // Enforce body size limit
+            const contentLength = req.headers.get('content-length');
+            if (contentLength !== null) {
+                const len = parseInt(contentLength, 10);
+                if (!Number.isFinite(len) || len > deps.maxBodySize) {
+                    return new Response(JSON.stringify({ error: 'Request body too large' }), { status: 413, headers: { 'Content-Type': 'application/json' } });
+                }
+            }
+            // Parse the request body to extract model name
+            const bodyText = await req.text();
+            if (new TextEncoder().encode(bodyText).byteLength > deps.maxBodySize) {
+                return new Response(JSON.stringify({ error: 'Request body too large' }), { status: 413, headers: { 'Content-Type': 'application/json' } });
+            }
+            let body;
+            try {
+                const parsed = JSON.parse(bodyText);
+                if (typeof parsed !== 'object' || parsed === null || Array.isArray(parsed)) {
+                    return new Response(JSON.stringify({ error: 'Request body must be a JSON object' }), { status: 400, headers: { 'Content-Type': 'application/json' } });
+                }
+                body = parsed;
+            }
+            catch {
+                return new Response(JSON.stringify({ error: 'Invalid JSON body' }), { status: 400, headers: { 'Content-Type': 'application/json' } });
+            }
+            const model = extractModel(body);
+            const pricePerThousand = resolveModelPrice(deps.pricing, model);
+            const isStreaming = body.stream === true;
+            // Inject stream_options for usage reporting if streaming
+            if (isStreaming && !body.stream_options) {
+                body.stream_options = { include_usage: true };
+            }
+            // Build upstream URL
+            const url = new URL(req.url);
+            const upstreamUrl = `${deps.upstream}${url.pathname}`;
+            // Fetch from upstream
+            const timeout = deps.upstreamTimeout ?? 120_000;
+            let upstreamRes;
+            try {
+                upstreamRes = await fetch(upstreamUrl, {
+                    method: 'POST',
+                    headers: { 'Content-Type': 'application/json' },
+                    body: JSON.stringify(body),
+                    signal: AbortSignal.timeout(timeout),
+                });
+            }
+            catch (err) {
+                // Upstream unreachable - refund estimated cost
+                if (paymentHash) {
+                    deps.reconcile(paymentHash, 0);
+                }
+                deps.logger?.error('upstream error', {
+                    endpoint: new URL(req.url).pathname,
+                    method: req.method,
+                    latencyMs: Date.now() - start,
+                    reason: err instanceof Error ? err.message : String(err),
+                });
+                return new Response(JSON.stringify({ error: 'Upstream inference API unreachable' }), { status: 502, headers: { 'Content-Type': 'application/json' } });
+            }
+            // If upstream returned an error, refund and return a generic error
+            // (don't forward raw upstream body — may leak internal details)
+            if (!upstreamRes.ok) {
+                if (paymentHash) {
+                    deps.reconcile(paymentHash, 0);
+                }
+                // Consume and discard the upstream error body to prevent connection leaks
+                await upstreamRes.body?.cancel().catch(() => { });
+                const status = upstreamRes.status >= 400 && upstreamRes.status < 600
+                    ? upstreamRes.status
+                    : 502;
+                return new Response(JSON.stringify({ error: `Upstream returned ${upstreamRes.status}` }), { status, headers: { 'Content-Type': 'application/json' } });
+            }
+            // Handle streaming response
+            if (isStreaming && upstreamRes.body) {
+                const { readable } = createStreamingProxy(upstreamRes.body, (tokenCount) => {
+                    // Release capacity slot when stream ends (not in finally)
+                    deps.capacity.release();
+                    if (!deps.flatPricing && paymentHash) {
+                        const satCost = tokenCostToSats(tokenCount, pricePerThousand);
+                        deps.reconcile(paymentHash, satCost);
+                    }
+                });
+                // Mark as streaming so finally doesn't double-release
+                streamingResponse = true;
+                return new Response(readable, {
+                    status: 200,
+                    headers: {
+                        'Content-Type': 'text/event-stream',
+                        'Cache-Control': 'no-cache',
+                        'Connection': 'keep-alive',
+                    },
+                });
+            }
+            // Handle non-streaming response — enforce size limit before parsing
+            const responseText = await upstreamRes.text();
+            if (new TextEncoder().encode(responseText).byteLength > deps.maxBodySize) {
+                if (paymentHash)
+                    deps.reconcile(paymentHash, 0);
+                return new Response(JSON.stringify({ error: 'Upstream response too large' }), { status: 502, headers: { 'Content-Type': 'application/json' } });
+            }
+            // eslint-disable-next-line @typescript-eslint/no-explicit-any
+            let responseBody;
+            try {
+                responseBody = JSON.parse(responseText);
+            }
+            catch {
+                return new Response(JSON.stringify({ error: 'Upstream returned invalid JSON' }), { status: 502, headers: { 'Content-Type': 'application/json' } });
+            }
+            const counter = new TokenCounter();
+            if (responseBody && typeof responseBody === 'object' && responseBody.usage) {
+                counter.setBufferedUsage(responseBody.usage);
+            }
+            const tokenCount = counter.finalCount();
+            const satCost = tokenCostToSats(tokenCount, pricePerThousand);
+            if (!deps.flatPricing && paymentHash) {
+                deps.reconcile(paymentHash, satCost);
+            }
+            return new Response(JSON.stringify(responseBody), {
+                status: 200,
+                headers: { 'Content-Type': 'application/json' },
+            });
+        }
+        finally {
+            // Streaming responses release capacity in the onComplete callback
+            if (!streamingResponse) {
+                deps.capacity.release();
+            }
+        }
+    };
+}
+//# sourceMappingURL=handler.js.map

package/dist/src/proxy/handler.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"handler.js","sourceRoot":"","sources":["../../../src/proxy/handler.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAA;AACjD,OAAO,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAA;AACrD,OAAO,EAAE,iBAAiB,EAAE,eAAe,EAAE,MAAM,cAAc,CAAA;AAKjE,kEAAkE;AAClE,MAAM,qBAAqB,GAAG,CAAC,sBAAsB,EAAE,iBAAiB,EAAE,gBAAgB,CAAC,CAAA;AAgB3F;;GAEG;AACH,SAAS,YAAY,CAAC,IAA6B;IACjD,OAAO,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAA;AACzD,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,kBAAkB,CAAC,IAAe;IAChD,OAAO,KAAK,UAAU,WAAW,CAC/B,GAAY,EACZ,WAA+B;QAE/B,gDAAgD;QAChD,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,UAAU,EAAE,EAAE,CAAC;YAChC,OAAO,IAAI,QAAQ,CACjB,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,sCAAsC,EAAE,CAAC,EACjE,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,aAAa,EAAE,GAAG,EAAE,EAAE,CACrF,CAAA;QACH,CAAC;QAED,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;QACxB,IAAI,iBAAiB,GAAG,KAAK,CAAA;QAC7B,IAAI,CAAC;YACH,uEAAuE;YACvE,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,QAAQ,CAAA;YAC7C,IAAI,CAAC,qBAAqB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,WAAW,KAAK,CAAC,CAAC,EAAE,CAAC;gBACxD,OAAO,IAAI,QAAQ,CACjB,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,CAAC,EACtC,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,EAAE,CACjE,CAAA;YACH,CAAC;YAED,wBAAwB;YACxB,MAAM,WAAW,GAAG,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,CAAA;YACnD,IAAI,CAAC,WAAW,IAAI,CAAC,WAAW,CAAC,QAAQ,CAAC,kBAAkB,CAAC,EAAE,CAAC;gBAC9D,OAAO,IAAI,QAAQ,CACjB,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,uCAAuC,EAAE,CAAC,EAClE,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,EAAE,CACjE,CAAA;YACH,CAAC;YAED,0BAA0B;YAC1B,MAAM,aAAa,GAAG,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAA;YACvD,IAAI,aAAa,KAAK,IAAI,EAAE,CAAC;gBAC3B,MAAM,GAAG,GAAG,QAAQ,CAAC,aAAa,EAAE,EAAE,CAAC,CAAA;gBACvC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,GAAG,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;oBACpD,OAAO,IAAI,QAAQ,CACjB,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,wBAAwB,EAAE,CAAC,EACnD,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,EAAE,CACjE,CAAA;gBACH,CAAC;YACH,CAAC;YAED,+CAA+C;YAC/C,MAAM,QAAQ,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAA;YACjC,IAAI,IAAI,WAAW,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,UAAU,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;gBACrE,OAAO,IAAI,QAAQ,CACjB,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,wBAAwB,EAAE,CAAC,EACnD,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,EAAE,CACjE,CAAA;YACH,CAAC;YAED,IAAI,IAA6B,CAAA;YACjC,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAA;gBACnC,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,KAAK,IAAI,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;oBAC3E,OAAO,IAAI,QAAQ,CACjB,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,oCAAoC,EAAE,CAAC,EAC/D,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,EAAE,CACjE,CAAA;gBACH,CAAC;gBACD,IAAI,GAAG,MAAM,CAAA;YACf,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,IAAI,QAAQ,CACjB,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,mBAAmB,EAAE,CAAC,EAC9C,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,EAAE,CACjE,CAAA;YACH,CAAC;YAED,MAAM,KAAK,GAAG,YAAY,CAAC,IAAI,CAAC,CAAA;YAChC,MAAM,gBAAgB,GAAG,iBAAiB,CAAC,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,CAAA;YAC/D,MAAM,WAAW,GAAG,IAAI,CAAC,MAAM,KAAK,IAAI,CAAA;YAExC,yDAAyD;YACzD,IAAI,WAAW,IAAI,CAAC,IAAI,CAAC,cAAc,EAAE,CAAC;gBACxC,IAAI,CAAC,cAAc,GAAG,EAAE,aAAa,EAAE,IAAI,EAAE,CAAA;YAC/C,CAAC;YAED,qBAAqB;YACrB,MAAM,GAAG,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;YAC5B,MAAM,WAAW,GAAG,GAAG,IAAI,CAAC,QAAQ,GAAG,GAAG,CAAC,QAAQ,EAAE,CAAA;YAErD,sBAAsB;YACtB,MAAM,OAAO,GAAG,IAAI,CAAC,eAAe,IAAI,OAAO,CAAA;YAC/C,IAAI,WAAqB,CAAA;YACzB,IAAI,CAAC;gBACH,WAAW,GAAG,MAAM,KAAK,CAAC,WAAW,EAAE;oBACrC,MAAM,EAAE,MAAM;oBACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;oBAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC;oBAC1B,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,OAAO,CAAC;iBACrC,CAAC,CAAA;YACJ,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,+CAA+C;gBAC/C,IAAI,WAAW,EAAE,CAAC;oBAChB,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC,CAAC,CAAA;gBAChC,CAAC;gBACD,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC,gBAAgB,EAAE;oBACnC,QAAQ,EAAE,IAAI,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,QAAQ;oBACnC,MAAM,EAAE,GAAG,CAAC,MAAM;oBAClB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;oBAC7B,MAAM,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;iBACzD,CAAC,CAAA;gBACF,OAAO,IAAI,QAAQ,CACjB,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,oCAAoC,EAAE,CAAC,EAC/D,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,EAAE,CACjE,CAAA;YACH,CAAC;YAED,mEAAmE;YACnE,gEAAgE;YAChE,IAAI,CAAC,WAAW,CAAC,EAAE,EAAE,CAAC;gBACpB,IAAI,WAAW,EAAE,CAAC;oBAChB,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC,CAAC,CAAA;gBAChC,CAAC;gBACD,0EAA0E;gBAC1E,MAAM,WAAW,CAAC,IAAI,EAAE,MAAM,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAA;gBAChD,MAAM,MAAM,GAAG,WAAW,CAAC,MAAM,IAAI,GAAG,IAAI,WAAW,CAAC,MAAM,GAAG,GAAG;oBAClE,CAAC,CAAC,WAAW,CAAC,MAAM;oBACpB,CAAC,CAAC,GAAG,CAAA;gBACP,OAAO,IAAI,QAAQ,CACjB,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,qBAAqB,WAAW,CAAC,MAAM,EAAE,EAAE,CAAC,EACpE,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,EAAE,CAC5D,CAAA;YACH,CAAC;YAED,4BAA4B;YAC5B,IAAI,WAAW,IAAI,WAAW,CAAC,IAAI,EAAE,CAAC;gBACpC,MAAM,EAAE,QAAQ,EAAE,GAAG,oBAAoB,CAAC,WAAW,CAAC,IAAI,EAAE,CAAC,UAAU,EAAE,EAAE;oBACzE,0DAA0D;oBAC1D,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,CAAA;oBACvB,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,WAAW,EAAE,CAAC;wBACrC,MAAM,OAAO,GAAG,eAAe,CAAC,UAAU,EAAE,gBAAgB,CAAC,CAAA;wBAC7D,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;oBACtC,CAAC;gBACH,CAAC,CAAC,CAAA;gBAEF,sDAAsD;gBACtD,iBAAiB,GAAG,IAAI,CAAA;gBAExB,OAAO,IAAI,QAAQ,CAAC,QAAQ,EAAE;oBAC5B,MAAM,EAAE,GAAG;oBACX,OAAO,EAAE;wBACP,cAAc,EAAE,mBAAmB;wBACnC,eAAe,EAAE,UAAU;wBAC3B,YAAY,EAAE,YAAY;qBAC3B;iBACF,CAAC,CAAA;YACJ,CAAC;YAED,oEAAoE;YACpE,MAAM,YAAY,GAAG,MAAM,WAAW,CAAC,IAAI,EAAE,CAAA;YAC7C,IAAI,IAAI,WAAW,EAAE,CAAC,MAAM,CAAC,YAAY,CAAC,CAAC,UAAU,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;gBACzE,IAAI,WAAW;oBAAE,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC,CAAC,CAAA;gBAC/C,OAAO,IAAI,QAAQ,CACjB,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,6BAA6B,EAAE,CAAC,EACxD,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,EAAE,CACjE,CAAA;YACH,CAAC;YACD,8DAA8D;YAC9D,IAAI,YAAiB,CAAA;YACrB,IAAI,CAAC;gBACH,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,CAAA;YACzC,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,IAAI,QAAQ,CACjB,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,gCAAgC,EAAE,CAAC,EAC3D,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE,EAAE,CACjE,CAAA;YACH,CAAC;YACD,MAAM,OAAO,GAAG,IAAI,YAAY,EAAE,CAAA;YAClC,IAAI,YAAY,IAAI,OAAO,YAAY,KAAK,QAAQ,IAAI,YAAY,CAAC,KAAK,EAAE,CAAC;gBAC3E,OAAO,CAAC,gBAAgB,CAAC,YAAY,CAAC,KAAK,CAAC,CAAA;YAC9C,CAAC;YACD,MAAM,UAAU,GAAG,OAAO,CAAC,UAAU,EAAE,CAAA;YACvC,MAAM,OAAO,GAAG,eAAe,CAAC,UAAU,EAAE,gBAAgB,CAAC,CAAA;YAE7D,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,WAAW,EAAE,CAAC;gBACrC,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;YACtC,CAAC;YAED,OAAO,IAAI,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC,YAAY,CAAC,EAAE;gBAChD,MAAM,EAAE,GAAG;gBACX,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;aAChD,CAAC,CAAA;QACJ,CAAC;gBAAS,CAAC;YACT,kEAAkE;YAClE,IAAI,CAAC,iBAAiB,EAAE,CAAC;gBACvB,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,CAAA;YACzB,CAAC;QACH,CAAC;IACH,CAAC,CAAA;AACH,CAAC"}

package/dist/src/proxy/pricing.d.ts

@@ -0,0 +1,17 @@
+import type { ModelPricing } from '../config.js';
+/**
+ * Resolves the price per 1k tokens for a model.
+ *
+ * Resolution order:
+ * 1. Exact match on model name
+ * 2. Case-insensitive match
+ * 3. Strip Ollama tag (model:tag -> model) and retry
+ * 4. Fall back to default price
+ */
+export declare function resolveModelPrice(pricing: ModelPricing, model: string): number;
+/**
+ * Converts a token count to a sat cost.
+ * Always rounds up (ceil) so the operator is never short-changed.
+ */
+export declare function tokenCostToSats(totalTokens: number, pricePerThousand: number): number;
+//# sourceMappingURL=pricing.d.ts.map

package/dist/src/proxy/pricing.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"pricing.d.ts","sourceRoot":"","sources":["../../../src/proxy/pricing.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,cAAc,CAAA;AAEhD;;;;;;;;GAQG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,YAAY,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,CAsB9E;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAAC,WAAW,EAAE,MAAM,EAAE,gBAAgB,EAAE,MAAM,GAAG,MAAM,CAGrF"}

package/dist/src/proxy/pricing.js

@@ -0,0 +1,42 @@
+/**
+ * Resolves the price per 1k tokens for a model.
+ *
+ * Resolution order:
+ * 1. Exact match on model name
+ * 2. Case-insensitive match
+ * 3. Strip Ollama tag (model:tag -> model) and retry
+ * 4. Fall back to default price
+ */
+export function resolveModelPrice(pricing, model) {
+    if (!model)
+        return pricing.default;
+    // Exact match
+    if (model in pricing.models)
+        return pricing.models[model];
+    // Case-insensitive match
+    const lower = model.toLowerCase();
+    for (const [key, value] of Object.entries(pricing.models)) {
+        if (key.toLowerCase() === lower)
+            return value;
+    }
+    // Strip Ollama tag (e.g. llama3:latest -> llama3)
+    const colonIdx = lower.indexOf(':');
+    if (colonIdx !== -1) {
+        const base = lower.slice(0, colonIdx);
+        for (const [key, value] of Object.entries(pricing.models)) {
+            if (key.toLowerCase() === base)
+                return value;
+        }
+    }
+    return pricing.default;
+}
+/**
+ * Converts a token count to a sat cost.
+ * Always rounds up (ceil) so the operator is never short-changed.
+ */
+export function tokenCostToSats(totalTokens, pricePerThousand) {
+    if (totalTokens <= 0)
+        return 0;
+    return Math.ceil(totalTokens * pricePerThousand / 1000);
+}
+//# sourceMappingURL=pricing.js.map

package/dist/src/proxy/pricing.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"pricing.js","sourceRoot":"","sources":["../../../src/proxy/pricing.ts"],"names":[],"mappings":"AAEA;;;;;;;;GAQG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAAqB,EAAE,KAAa;IACpE,IAAI,CAAC,KAAK;QAAE,OAAO,OAAO,CAAC,OAAO,CAAA;IAElC,cAAc;IACd,IAAI,KAAK,IAAI,OAAO,CAAC,MAAM;QAAE,OAAO,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,CAAA;IAEzD,yBAAyB;IACzB,MAAM,KAAK,GAAG,KAAK,CAAC,WAAW,EAAE,CAAA;IACjC,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;QAC1D,IAAI,GAAG,CAAC,WAAW,EAAE,KAAK,KAAK;YAAE,OAAO,KAAK,CAAA;IAC/C,CAAC;IAED,kDAAkD;IAClD,MAAM,QAAQ,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA;IACnC,IAAI,QAAQ,KAAK,CAAC,CAAC,EAAE,CAAC;QACpB,MAAM,IAAI,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAA;QACrC,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;YAC1D,IAAI,GAAG,CAAC,WAAW,EAAE,KAAK,IAAI;gBAAE,OAAO,KAAK,CAAA;QAC9C,CAAC;IACH,CAAC;IAED,OAAO,OAAO,CAAC,OAAO,CAAA;AACxB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,WAAmB,EAAE,gBAAwB;IAC3E,IAAI,WAAW,IAAI,CAAC;QAAE,OAAO,CAAC,CAAA;IAC9B,OAAO,IAAI,CAAC,IAAI,CAAC,WAAW,GAAG,gBAAgB,GAAG,IAAI,CAAC,CAAA;AACzD,CAAC"}

package/dist/src/proxy/streaming.d.ts

@@ -0,0 +1,12 @@
+/**
+ * Creates a ReadableStream that pipes upstream SSE chunks through while counting tokens.
+ *
+ * @param upstream - The upstream SSE ReadableStream
+ * @param onComplete - Called with the final token count after the stream ends or errors
+ * @param inactivityTimeoutMs - Max time between chunks before aborting (default: 120s)
+ * @returns An object with the readable side
+ */
+export declare function createStreamingProxy(upstream: ReadableStream<Uint8Array>, onComplete: (tokenCount: number) => void, inactivityTimeoutMs?: number): {
+    readable: ReadableStream<Uint8Array>;
+};
+//# sourceMappingURL=streaming.d.ts.map

package/dist/src/proxy/streaming.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"streaming.d.ts","sourceRoot":"","sources":["../../../src/proxy/streaming.ts"],"names":[],"mappings":"AAKA;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAClC,QAAQ,EAAE,cAAc,CAAC,UAAU,CAAC,EACpC,UAAU,EAAE,CAAC,UAAU,EAAE,MAAM,KAAK,IAAI,EACxC,mBAAmB,GAAE,MAAsC,GAC1D;IAAE,QAAQ,EAAE,cAAc,CAAC,UAAU,CAAC,CAAA;CAAE,CAsD1C"}

package/dist/src/proxy/streaming.js

@@ -0,0 +1,68 @@
+import { TokenCounter } from './token-counter.js';
+/** Default inactivity timeout for streaming responses (2 minutes). */
+const DEFAULT_INACTIVITY_TIMEOUT_MS = 120_000;
+/**
+ * Creates a ReadableStream that pipes upstream SSE chunks through while counting tokens.
+ *
+ * @param upstream - The upstream SSE ReadableStream
+ * @param onComplete - Called with the final token count after the stream ends or errors
+ * @param inactivityTimeoutMs - Max time between chunks before aborting (default: 120s)
+ * @returns An object with the readable side
+ */
+export function createStreamingProxy(upstream, onComplete, inactivityTimeoutMs = DEFAULT_INACTIVITY_TIMEOUT_MS) {
+    const counter = new TokenCounter();
+    const decoder = new TextDecoder();
+    let completeCalled = false;
+    let inactivityTimer;
+    function finish() {
+        if (completeCalled)
+            return;
+        completeCalled = true;
+        clearTimeout(inactivityTimer);
+        onComplete(counter.finalCount());
+    }
+    const readable = new ReadableStream({
+        async start(controller) {
+            const reader = upstream.getReader();
+            function resetTimer() {
+                clearTimeout(inactivityTimer);
+                inactivityTimer = setTimeout(() => {
+                    reader.cancel('inactivity timeout').catch(() => { });
+                    controller.close();
+                    finish();
+                }, inactivityTimeoutMs);
+            }
+            resetTimer();
+            try {
+                while (true) {
+                    const { done, value } = await reader.read();
+                    if (done)
+                        break;
+                    resetTimer();
+                    controller.enqueue(value);
+                    const text = decoder.decode(value, { stream: true });
+                    counter.ingestSSEChunk(text);
+                }
+                controller.close();
+            }
+            catch {
+                // Upstream errored — close gracefully
+                try {
+                    controller.close();
+                }
+                catch { /* already closed */ }
+            }
+            finally {
+                clearTimeout(inactivityTimer);
+                finish();
+            }
+        },
+        cancel() {
+            // Client disconnected
+            clearTimeout(inactivityTimer);
+            finish();
+        },
+    });
+    return { readable };
+}
+//# sourceMappingURL=streaming.js.map

package/dist/src/proxy/streaming.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"streaming.js","sourceRoot":"","sources":["../../../src/proxy/streaming.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAA;AAEjD,sEAAsE;AACtE,MAAM,6BAA6B,GAAG,OAAO,CAAA;AAE7C;;;;;;;GAOG;AACH,MAAM,UAAU,oBAAoB,CAClC,QAAoC,EACpC,UAAwC,EACxC,sBAA8B,6BAA6B;IAE3D,MAAM,OAAO,GAAG,IAAI,YAAY,EAAE,CAAA;IAClC,MAAM,OAAO,GAAG,IAAI,WAAW,EAAE,CAAA;IACjC,IAAI,cAAc,GAAG,KAAK,CAAA;IAC1B,IAAI,eAA0D,CAAA;IAE9D,SAAS,MAAM;QACb,IAAI,cAAc;YAAE,OAAM;QAC1B,cAAc,GAAG,IAAI,CAAA;QACrB,YAAY,CAAC,eAAe,CAAC,CAAA;QAC7B,UAAU,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC,CAAA;IAClC,CAAC;IAED,MAAM,QAAQ,GAAG,IAAI,cAAc,CAAa;QAC9C,KAAK,CAAC,KAAK,CAAC,UAAU;YACpB,MAAM,MAAM,GAAG,QAAQ,CAAC,SAAS,EAAE,CAAA;YAEnC,SAAS,UAAU;gBACjB,YAAY,CAAC,eAAe,CAAC,CAAA;gBAC7B,eAAe,GAAG,UAAU,CAAC,GAAG,EAAE;oBAChC,MAAM,CAAC,MAAM,CAAC,oBAAoB,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAA;oBACnD,UAAU,CAAC,KAAK,EAAE,CAAA;oBAClB,MAAM,EAAE,CAAA;gBACV,CAAC,EAAE,mBAAmB,CAAC,CAAA;YACzB,CAAC;YAED,UAAU,EAAE,CAAA;YAEZ,IAAI,CAAC;gBACH,OAAO,IAAI,EAAE,CAAC;oBACZ,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,MAAM,MAAM,CAAC,IAAI,EAAE,CAAA;oBAC3C,IAAI,IAAI;wBAAE,MAAK;oBACf,UAAU,EAAE,CAAA;oBACZ,UAAU,CAAC,OAAO,CAAC,KAAK,CAAC,CAAA;oBACzB,MAAM,IAAI,GAAG,OAAO,CAAC,MAAM,CAAC,KAAK,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,CAAA;oBACpD,OAAO,CAAC,cAAc,CAAC,IAAI,CAAC,CAAA;gBAC9B,CAAC;gBACD,UAAU,CAAC,KAAK,EAAE,CAAA;YACpB,CAAC;YAAC,MAAM,CAAC;gBACP,sCAAsC;gBACtC,IAAI,CAAC;oBAAC,UAAU,CAAC,KAAK,EAAE,CAAA;gBAAC,CAAC;gBAAC,MAAM,CAAC,CAAC,oBAAoB,CAAC,CAAC;YAC3D,CAAC;oBAAS,CAAC;gBACT,YAAY,CAAC,eAAe,CAAC,CAAA;gBAC7B,MAAM,EAAE,CAAA;YACV,CAAC;QACH,CAAC;QACD,MAAM;YACJ,sBAAsB;YACtB,YAAY,CAAC,eAAe,CAAC,CAAA;YAC7B,MAAM,EAAE,CAAA;QACV,CAAC;KACF,CAAC,CAAA;IAEF,OAAO,EAAE,QAAQ,EAAE,CAAA;AACrB,CAAC"}

package/dist/src/proxy/token-counter.d.ts

@@ -0,0 +1,22 @@
+/**
+ * Counts tokens from OpenAI-compatible responses.
+ *
+ * Priority:
+ * 1. Buffered usage (from non-streaming JSON response)
+ * 2. Usage from final SSE chunk (stream_options: { include_usage: true })
+ * 3. Content chunk count (fallback - 1 chunk ~= 1 token)
+ */
+export declare class TokenCounter {
+    private bufferedUsage;
+    private sseUsage;
+    private contentChunkCount;
+    /** Set usage from a buffered (non-streaming) JSON response. */
+    setBufferedUsage(usage: Record<string, unknown>): void;
+    /** Ingest an SSE chunk (may contain multiple events). */
+    ingestSSEChunk(chunk: string): void;
+    /** Returns the final token count using the best available source.
+     *  Uses prompt_tokens from usage stats + content chunk count for completion.
+     *  This avoids billing for reasoning/thinking tokens that some models produce. */
+    finalCount(): number;
+}
+//# sourceMappingURL=token-counter.d.ts.map

package/dist/src/proxy/token-counter.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"token-counter.d.ts","sourceRoot":"","sources":["../../../src/proxy/token-counter.ts"],"names":[],"mappings":"AAMA;;;;;;;GAOG;AACH,qBAAa,YAAY;IACvB,OAAO,CAAC,aAAa,CAAyB;IAC9C,OAAO,CAAC,QAAQ,CAAyB;IACzC,OAAO,CAAC,iBAAiB,CAAI;IAE7B,+DAA+D;IAC/D,gBAAgB,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI;IAQtD,yDAAyD;IACzD,cAAc,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI;IAkCnC;;sFAEkF;IAClF,UAAU,IAAI,MAAM;CAQrB"}

package/dist/src/proxy/token-counter.js

@@ -0,0 +1,66 @@
+/**
+ * Counts tokens from OpenAI-compatible responses.
+ *
+ * Priority:
+ * 1. Buffered usage (from non-streaming JSON response)
+ * 2. Usage from final SSE chunk (stream_options: { include_usage: true })
+ * 3. Content chunk count (fallback - 1 chunk ~= 1 token)
+ */
+export class TokenCounter {
+    bufferedUsage = null;
+    sseUsage = null;
+    contentChunkCount = 0;
+    /** Set usage from a buffered (non-streaming) JSON response. */
+    setBufferedUsage(usage) {
+        this.bufferedUsage = {
+            prompt_tokens: typeof usage.prompt_tokens === 'number' ? usage.prompt_tokens : undefined,
+            completion_tokens: typeof usage.completion_tokens === 'number' ? usage.completion_tokens : undefined,
+            total_tokens: typeof usage.total_tokens === 'number' ? usage.total_tokens : undefined,
+        };
+    }
+    /** Ingest an SSE chunk (may contain multiple events). */
+    ingestSSEChunk(chunk) {
+        const lines = chunk.split('\n');
+        for (const line of lines) {
+            if (!line.startsWith('data: '))
+                continue;
+            const data = line.slice(6).trim();
+            if (data === '[DONE]')
+                continue;
+            try {
+                const parsed = JSON.parse(data);
+                // Check for usage in this chunk
+                if (parsed.usage) {
+                    this.sseUsage = {
+                        prompt_tokens: parsed.usage.prompt_tokens,
+                        completion_tokens: parsed.usage.completion_tokens,
+                        total_tokens: parsed.usage.total_tokens,
+                    };
+                }
+                // Count content chunks
+                const choices = parsed.choices;
+                if (Array.isArray(choices)) {
+                    for (const choice of choices) {
+                        if (choice.delta?.content !== undefined && choice.delta.content !== '') {
+                            this.contentChunkCount++;
+                        }
+                    }
+                }
+            }
+            catch {
+                // Malformed JSON - skip
+            }
+        }
+    }
+    /** Returns the final token count using the best available source.
+     *  Uses prompt_tokens from usage stats + content chunk count for completion.
+     *  This avoids billing for reasoning/thinking tokens that some models produce. */
+    finalCount() {
+        const usage = this.bufferedUsage ?? this.sseUsage;
+        const promptTokens = usage?.prompt_tokens ?? 0;
+        // Content chunk count is the most reliable measure of actual output
+        // since it excludes reasoning tokens that models like qwen3 produce
+        return promptTokens + this.contentChunkCount;
+    }
+}
+//# sourceMappingURL=token-counter.js.map

package/dist/src/proxy/token-counter.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"token-counter.js","sourceRoot":"","sources":["../../../src/proxy/token-counter.ts"],"names":[],"mappings":"AAMA;;;;;;;GAOG;AACH,MAAM,OAAO,YAAY;IACf,aAAa,GAAqB,IAAI,CAAA;IACtC,QAAQ,GAAqB,IAAI,CAAA;IACjC,iBAAiB,GAAG,CAAC,CAAA;IAE7B,+DAA+D;IAC/D,gBAAgB,CAAC,KAA8B;QAC7C,IAAI,CAAC,aAAa,GAAG;YACnB,aAAa,EAAE,OAAO,KAAK,CAAC,aAAa,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS;YACxF,iBAAiB,EAAE,OAAO,KAAK,CAAC,iBAAiB,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,iBAAiB,CAAC,CAAC,CAAC,SAAS;YACpG,YAAY,EAAE,OAAO,KAAK,CAAC,YAAY,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,CAAC,SAAS;SACtF,CAAA;IACH,CAAC;IAED,yDAAyD;IACzD,cAAc,CAAC,KAAa;QAC1B,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,CAAA;QAC/B,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC;gBAAE,SAAQ;YACxC,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAA;YACjC,IAAI,IAAI,KAAK,QAAQ;gBAAE,SAAQ;YAE/B,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAA;gBAE/B,gCAAgC;gBAChC,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;oBACjB,IAAI,CAAC,QAAQ,GAAG;wBACd,aAAa,EAAE,MAAM,CAAC,KAAK,CAAC,aAAa;wBACzC,iBAAiB,EAAE,MAAM,CAAC,KAAK,CAAC,iBAAiB;wBACjD,YAAY,EAAE,MAAM,CAAC,KAAK,CAAC,YAAY;qBACxC,CAAA;gBACH,CAAC;gBAED,uBAAuB;gBACvB,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,CAAA;gBAC9B,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;oBAC3B,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;wBAC7B,IAAI,MAAM,CAAC,KAAK,EAAE,OAAO,KAAK,SAAS,IAAI,MAAM,CAAC,KAAK,CAAC,OAAO,KAAK,EAAE,EAAE,CAAC;4BACvE,IAAI,CAAC,iBAAiB,EAAE,CAAA;wBAC1B,CAAC;oBACH,CAAC;gBACH,CAAC;YACH,CAAC;YAAC,MAAM,CAAC;gBACP,wBAAwB;YAC1B,CAAC;QACH,CAAC;IACH,CAAC;IAED;;sFAEkF;IAClF,UAAU;QACR,MAAM,KAAK,GAAG,IAAI,CAAC,aAAa,IAAI,IAAI,CAAC,QAAQ,CAAA;QACjD,MAAM,YAAY,GAAG,KAAK,EAAE,aAAa,IAAI,CAAC,CAAA;QAE9C,oEAAoE;QACpE,oEAAoE;QACpE,OAAO,YAAY,GAAG,IAAI,CAAC,iBAAiB,CAAA;IAC9C,CAAC;CACF"}

package/dist/src/server.d.ts

@@ -0,0 +1,9 @@
+import { Hono } from 'hono';
+import type { TollBoothEnv } from '@thecryptodonkey/toll-booth/hono';
+import type { TokenTollConfig } from './config.js';
+export interface TokenTollServer {
+    app: Hono<TollBoothEnv>;
+    close: () => void;
+}
+export declare function createTokenTollServer(config: TokenTollConfig): TokenTollServer;
+//# sourceMappingURL=server.d.ts.map

package/dist/src/server.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/server.ts"],"names":[],"mappings":"AAGA,OAAO,EAAE,IAAI,EAAE,MAAM,MAAM,CAAA;AAU3B,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,kCAAkC,CAAA;AACpE,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAA;AAUlD,MAAM,WAAW,eAAe;IAC9B,GAAG,EAAE,IAAI,CAAC,YAAY,CAAC,CAAA;IACvB,KAAK,EAAE,MAAM,IAAI,CAAA;CAClB;AAgBD,wBAAgB,qBAAqB,CAAC,MAAM,EAAE,eAAe,GAAG,eAAe,CAqO9E"}