@relay-harness/coding-agent 0.0.0-stage → 1.0.2

package/dist/extensions/llama/index.js

@@ -0,0 +1,219 @@
+import { formatBytes, LlamaClient, normalizeLlamaServerUrl } from "./client.js";
+import { findHuggingFaceToken, HuggingFaceClient } from "./huggingface.js";
+import { createLlamaProvider, LLAMA_PROVIDER_ID } from "./provider.js";
+import { runWithProgress, showLlamaUi } from "./ui.js";
+function modelIsLoaded(model) {
+    return model.status.value === "loaded" || model.status.value === "sleeping";
+}
+function isConnectionError(error) {
+    if (!(error instanceof Error))
+        return false;
+    const message = `${error.name} ${error.message}`.toLowerCase();
+    return message.includes("fetch failed") || message.includes("timeout") || message.includes("network");
+}
+function connectionErrorMessage(error) {
+    if (isConnectionError(error))
+        return "Could not connect to the server.";
+    return error instanceof Error ? error.message : String(error);
+}
+function parseHuggingFaceModel(value) {
+    const colon = value.indexOf(":", value.indexOf("/") + 1);
+    return colon < 0
+        ? { repository: value }
+        : { repository: value.slice(0, colon), quantization: value.slice(colon + 1) };
+}
+async function configuredClient(ctx) {
+    const result = await ctx.modelRegistry.getProviderAuth(LLAMA_PROVIDER_ID);
+    if (!result) {
+        ctx.ui.notify(`Configure llama.cpp with /login ${LLAMA_PROVIDER_ID}`, "warning");
+        return undefined;
+    }
+    const configuredUrl = result.env?.LLAMA_BASE_URL;
+    const serverUrl = normalizeLlamaServerUrl(typeof configuredUrl === "string" && configuredUrl ? configuredUrl : (result.auth.baseUrl ?? ""));
+    return new LlamaClient(serverUrl, result.auth.apiKey);
+}
+export default function llamaExtension(relay) {
+    const provider = createLlamaProvider();
+    relay.registerProvider(provider.provider);
+    const syncCatalog = async (ctx, client, catalog) => {
+        const signal = AbortSignal.timeout(15_000);
+        const current = catalog ?? (await client.list({ signal }));
+        provider.setCatalog(current, client.serverUrl);
+        const result = await ctx.modelRegistry.refresh({
+            providers: [LLAMA_PROVIDER_ID],
+            // /llama already contacted the configured llama.cpp server, so keep this refresh live even in RELAY_OFFLINE.
+            allowNetwork: true,
+            signal,
+        });
+        if (result.aborted)
+            throw new Error("Model catalog refresh timed out.");
+        const refreshError = result.errors.get(LLAMA_PROVIDER_ID);
+        if (refreshError)
+            throw refreshError;
+        return current;
+    };
+    const loadModel = async (ctx, ui, client, catalog, target) => {
+        const loaded = catalog.filter((model) => model.id !== target.id && modelIsLoaded(model));
+        let replace = false;
+        if (loaded.length > 0) {
+            const choice = await ui.select(`${loaded.length} model${loaded.length === 1 ? " is" : "s are"} loaded`, [
+                "Unload all and load",
+                "Keep loaded and load",
+                "Cancel",
+            ]);
+            if (!choice || choice === "Cancel")
+                return;
+            replace = choice === "Unload all and load";
+        }
+        const restoreLoaded = async () => {
+            ctx.ui.notify("Restoring previously loaded models");
+            for (const model of loaded)
+                await client.loadAndWait(model.id, () => { });
+            await syncCatalog(ctx, client);
+        };
+        if (replace) {
+            for (const model of loaded)
+                await client.unloadAndWait(model.id);
+        }
+        try {
+            const result = await runWithProgress(ui, {
+                title: "Loading model",
+                model: target.id,
+                initialMessage: "Starting…",
+                cancelTitle: "Stop loading?",
+                cancelMessage: target.id,
+                run: (signal, update) => client.loadAndWait(target.id, update, signal),
+                cancel: () => client.unload(target.id),
+            });
+            if (result.cancelled) {
+                if (replace)
+                    await restoreLoaded();
+                return;
+            }
+            const refreshed = await syncCatalog(ctx, client);
+            const loadedModel = refreshed.find((model) => model.id === target.id);
+            ctx.ui.notify(loadedModel?.status.value === "loaded" ? `Loaded ${target.id}` : `Load started for ${target.id}`);
+        }
+        catch (error) {
+            if (replace) {
+                try {
+                    await restoreLoaded();
+                }
+                catch {
+                    // Preserve the original load error.
+                }
+            }
+            throw error;
+        }
+    };
+    const unloadModel = async (ctx, ui, client, model) => {
+        if (!(await ui.confirm("Unload model?", model.id)))
+            return;
+        await client.unloadAndWait(model.id);
+        await syncCatalog(ctx, client);
+        ctx.ui.notify(`Unloaded ${model.id}`);
+    };
+    const downloadModel = async (ctx, ui, client) => {
+        const huggingFace = new HuggingFaceClient(await findHuggingFaceToken());
+        const selected = await ui.searchModels((query, signal) => huggingFace.search(query, signal));
+        if (!selected)
+            return;
+        const parsed = parseHuggingFaceModel(selected);
+        ui.showStatus("Loading model details", parsed.repository);
+        const details = await huggingFace.details(parsed.repository);
+        if (details.gated) {
+            const approval = details.gated === "manual" ? "Manual approval is required" : "Accept the access terms";
+            const choice = await ui.select(`Hugging Face access required\n${details.id}\n\n${approval} at:\nhttps://huggingface.co/${details.id}\n\nThe llama.cpp server needs HF_TOKEN with access.`, ["Continue", "Back"]);
+            if (choice !== "Continue")
+                return;
+        }
+        let quantization = parsed.quantization;
+        if (!quantization && details.quantizations.length > 0) {
+            const options = details.quantizations.map((entry) => {
+                const detail = [
+                    entry.size === undefined ? undefined : formatBytes(entry.size),
+                    entry.name === "Q4_K_M" ? "recommended" : undefined,
+                ]
+                    .filter((value) => Boolean(value))
+                    .join(" · ");
+                return detail ? `${entry.name} · ${detail}` : entry.name;
+            });
+            const choice = await ui.select(`Select quantization\n${details.id}`, options);
+            if (!choice)
+                return;
+            quantization = details.quantizations[options.indexOf(choice)]?.name;
+            if (!quantization)
+                return;
+        }
+        const model = quantization ? `${details.id}:${quantization}` : details.id;
+        const result = await runWithProgress(ui, {
+            title: "Downloading model",
+            model,
+            initialMessage: "Starting…",
+            cancelTitle: "Stop download?",
+            cancelMessage: model,
+            run: (signal, update) => client.downloadAndWait(model, update, signal),
+            cancel: () => client.unload(model),
+        });
+        if (result.cancelled)
+            return;
+        await syncCatalog(ctx, client, result.value);
+        ctx.ui.notify(`Downloaded ${model}`);
+    };
+    relay.registerCommand("llama", {
+        description: "Manage llama.cpp router models",
+        handler: async (_args, ctx) => {
+            if (ctx.mode !== "tui") {
+                ctx.ui.notify("/llama is available in interactive mode", "warning");
+                return;
+            }
+            const client = await configuredClient(ctx);
+            if (!client)
+                return;
+            await showLlamaUi(ctx, async (ui) => {
+                const readCatalog = async () => {
+                    while (true) {
+                        try {
+                            return await syncCatalog(ctx, client);
+                        }
+                        catch (error) {
+                            if ((await ui.connectionError(client.serverUrl, connectionErrorMessage(error))) === "close") {
+                                return undefined;
+                            }
+                        }
+                    }
+                };
+                let catalog = await readCatalog();
+                if (!catalog)
+                    return;
+                while (true) {
+                    const action = await ui.showModels(client.serverUrl, catalog);
+                    if (action.type === "close")
+                        return;
+                    let actionError;
+                    try {
+                        if (action.type === "download")
+                            await downloadModel(ctx, ui, client);
+                        else if (modelIsLoaded(action.model))
+                            await unloadModel(ctx, ui, client, action.model);
+                        else if (action.model.status.value === "unloaded")
+                            await loadModel(ctx, ui, client, catalog, action.model);
+                        else
+                            ctx.ui.notify(`${action.model.id} is ${action.model.status.value}`, "warning");
+                    }
+                    catch (error) {
+                        actionError = error;
+                    }
+                    const refreshed = await readCatalog();
+                    if (!refreshed)
+                        return;
+                    catalog = refreshed;
+                    if (actionError && !isConnectionError(actionError)) {
+                        ctx.ui.notify(actionError instanceof Error ? actionError.message : String(actionError), "error");
+                    }
+                }
+            });
+        },
+    });
+}
+//# sourceMappingURL=index.js.map

package/dist/extensions/llama/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/extensions/llama/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,WAAW,EAAuB,uBAAuB,EAAE,MAAM,aAAa,CAAC;AACrG,OAAO,EAAE,oBAAoB,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAC3E,OAAO,EAAE,mBAAmB,EAAE,iBAAiB,EAAE,MAAM,eAAe,CAAC;AACvE,OAAO,EAAgB,eAAe,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AAErE,SAAS,aAAa,CAAC,KAAqB;IAC3C,OAAO,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,CAAC;AAC7E,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAc;IACxC,IAAI,CAAC,CAAC,KAAK,YAAY,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC5C,MAAM,OAAO,GAAG,GAAG,KAAK,CAAC,IAAI,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC,WAAW,EAAE,CAAC;IAC/D,OAAO,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;AACvG,CAAC;AAED,SAAS,sBAAsB,CAAC,KAAc;IAC7C,IAAI,iBAAiB,CAAC,KAAK,CAAC;QAAE,OAAO,kCAAkC,CAAC;IACxE,OAAO,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AAC/D,CAAC;AAED,SAAS,qBAAqB,CAAC,KAAa;IAC3C,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,EAAE,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;IACzD,OAAO,KAAK,GAAG,CAAC;QACf,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,EAAE;QACvB,CAAC,CAAC,EAAE,UAAU,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,YAAY,EAAE,KAAK,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC,EAAE,CAAC;AAChF,CAAC;AAED,KAAK,UAAU,gBAAgB,CAAC,GAA4B;IAC3D,MAAM,MAAM,GAAG,MAAM,GAAG,CAAC,aAAa,CAAC,eAAe,CAAC,iBAAiB,CAAC,CAAC;IAC1E,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,mCAAmC,iBAAiB,EAAE,EAAE,SAAS,CAAC,CAAC;QACjF,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,MAAM,aAAa,GAAG,MAAM,CAAC,GAAG,EAAE,cAAc,CAAC;IACjD,MAAM,SAAS,GAAG,uBAAuB,CACxC,OAAO,aAAa,KAAK,QAAQ,IAAI,aAAa,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CAAC,CAChG,CAAC;IACF,OAAO,IAAI,WAAW,CAAC,SAAS,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AACvD,CAAC;AAED,MAAM,CAAC,OAAO,UAAU,cAAc,CAAC,KAAmB;IACzD,MAAM,QAAQ,GAAG,mBAAmB,EAAE,CAAC;IACvC,KAAK,CAAC,gBAAgB,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAE1C,MAAM,WAAW,GAAG,KAAK,EACxB,GAA4B,EAC5B,MAAmB,EACnB,OAA0B,EACE,EAAE;QAC9B,MAAM,MAAM,GAAG,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC3C,MAAM,OAAO,GAAG,OAAO,IAAI,CAAC,MAAM,MAAM,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC;QAC3D,QAAQ,CAAC,UAAU,CAAC,OAAO,EAAE,MAAM,CAAC,SAAS,CAAC,CAAC;QAC/C,MAAM,MAAM,GAAG,MAAM,GAAG,CAAC,aAAa,CAAC,OAAO,CAAC;YAC9C,SAAS,EAAE,CAAC,iBAAiB,CAAC;YAC9B,6GAA6G;YAC7G,YAAY,EAAE,IAAI;YAClB,MAAM;SACN,CAAC,CAAC;QACH,IAAI,MAAM,CAAC,OAAO;YAAE,MAAM,IAAI,KAAK,CAAC,kCAAkC,CAAC,CAAC;QACxE,MAAM,YAAY,GAAG,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,iBAAiB,CAAC,CAAC;QAC1D,IAAI,YAAY;YAAE,MAAM,YAAY,CAAC;QACrC,OAAO,OAAO,CAAC;IAChB,CAAC,CAAC;IAEF,MAAM,SAAS,GAAG,KAAK,EACtB,GAA4B,EAC5B,EAAW,EACX,MAAmB,EACnB,OAAyB,EACzB,MAAsB,EACN,EAAE;QAClB,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,IAAI,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC;QACzF,IAAI,OAAO,GAAG,KAAK,CAAC;QACpB,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACvB,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,SAAS,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,SAAS,EAAE;gBACvG,qBAAqB;gBACrB,sBAAsB;gBACtB,QAAQ;aACR,CAAC,CAAC;YACH,IAAI,CAAC,MAAM,IAAI,MAAM,KAAK,QAAQ;gBAAE,OAAO;YAC3C,OAAO,GAAG,MAAM,KAAK,qBAAqB,CAAC;QAC5C,CAAC;QAED,MAAM,aAAa,GAAG,KAAK,IAAmB,EAAE;YAC/C,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,oCAAoC,CAAC,CAAC;YACpD,KAAK,MAAM,KAAK,IAAI,MAAM;gBAAE,MAAM,MAAM,CAAC,WAAW,CAAC,KAAK,CAAC,EAAE,EAAE,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;YACzE,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAChC,CAAC,CAAC;QACF,IAAI,OAAO,EAAE,CAAC;YACb,KAAK,MAAM,KAAK,IAAI,MAAM;gBAAE,MAAM,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAClE,CAAC;QAED,IAAI,CAAC;YACJ,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,EAAE,EAAE;gBACxC,KAAK,EAAE,eAAe;gBACtB,KAAK,EAAE,MAAM,CAAC,EAAE;gBAChB,cAAc,EAAE,WAAW;gBAC3B,WAAW,EAAE,eAAe;gBAC5B,aAAa,EAAE,MAAM,CAAC,EAAE;gBACxB,GAAG,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC;gBACtE,MAAM,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC;aACtC,CAAC,CAAC;YACH,IAAI,MAAM,CAAC,SAAS,EAAE,CAAC;gBACtB,IAAI,OAAO;oBAAE,MAAM,aAAa,EAAE,CAAC;gBACnC,OAAO;YACR,CAAC;YACD,MAAM,SAAS,GAAG,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YACjD,MAAM,WAAW,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,MAAM,CAAC,EAAE,CAAC,CAAC;YACtE,GAAG,CAAC,EAAE,CAAC,MAAM,CACZ,WAAW,EAAE,MAAM,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,UAAU,MAAM,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,oBAAoB,MAAM,CAAC,EAAE,EAAE,CAChG,CAAC;QACH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,IAAI,OAAO,EAAE,CAAC;gBACb,IAAI,CAAC;oBACJ,MAAM,aAAa,EAAE,CAAC;gBACvB,CAAC;gBAAC,MAAM,CAAC;oBACR,oCAAoC;gBACrC,CAAC;YACF,CAAC;YACD,MAAM,KAAK,CAAC;QACb,CAAC;IACF,CAAC,CAAC;IAEF,MAAM,WAAW,GAAG,KAAK,EACxB,GAA4B,EAC5B,EAAW,EACX,MAAmB,EACnB,KAAqB,EACL,EAAE;QAClB,IAAI,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,eAAe,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;YAAE,OAAO;QAC3D,MAAM,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACrC,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAC/B,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,YAAY,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC;IACvC,CAAC,CAAC;IAEF,MAAM,aAAa,GAAG,KAAK,EAAE,GAA4B,EAAE,EAAW,EAAE,MAAmB,EAAiB,EAAE;QAC7G,MAAM,WAAW,GAAG,IAAI,iBAAiB,CAAC,MAAM,oBAAoB,EAAE,CAAC,CAAC;QACxE,MAAM,QAAQ,GAAG,MAAM,EAAE,CAAC,YAAY,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,WAAW,CAAC,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC;QAC7F,IAAI,CAAC,QAAQ;YAAE,OAAO;QACtB,MAAM,MAAM,GAAG,qBAAqB,CAAC,QAAQ,CAAC,CAAC;QAC/C,EAAE,CAAC,UAAU,CAAC,uBAAuB,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;QAC1D,MAAM,OAAO,GAAG,MAAM,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;QAC7D,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;YACnB,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,6BAA6B,CAAC,CAAC,CAAC,yBAAyB,CAAC;YACxG,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAC7B,iCAAiC,OAAO,CAAC,EAAE,OAAO,QAAQ,gCAAgC,OAAO,CAAC,EAAE,sDAAsD,EAC1J,CAAC,UAAU,EAAE,MAAM,CAAC,CACpB,CAAC;YACF,IAAI,MAAM,KAAK,UAAU;gBAAE,OAAO;QACnC,CAAC;QACD,IAAI,YAAY,GAAG,MAAM,CAAC,YAAY,CAAC;QACvC,IAAI,CAAC,YAAY,IAAI,OAAO,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACvD,MAAM,OAAO,GAAG,OAAO,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;gBACnD,MAAM,MAAM,GAAG;oBACd,KAAK,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,KAAK,CAAC,IAAI,CAAC;oBAC9D,KAAK,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS;iBACnD;qBACC,MAAM,CAAC,CAAC,KAAK,EAAmB,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;qBAClD,IAAI,CAAC,KAAK,CAAC,CAAC;gBACd,OAAO,MAAM,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,IAAI,MAAM,MAAM,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC;YAC1D,CAAC,CAAC,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,MAAM,CAAC,wBAAwB,OAAO,CAAC,EAAE,EAAE,EAAE,OAAO,CAAC,CAAC;YAC9E,IAAI,CAAC,MAAM;gBAAE,OAAO;YACpB,YAAY,GAAG,OAAO,CAAC,aAAa,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,EAAE,IAAI,CAAC;YACpE,IAAI,CAAC,YAAY;gBAAE,OAAO;QAC3B,CAAC;QACD,MAAM,KAAK,GAAG,YAAY,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,EAAE,IAAI,YAAY,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;QAC1E,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,EAAE,EAAE;YACxC,KAAK,EAAE,mBAAmB;YAC1B,KAAK;YACL,cAAc,EAAE,WAAW;YAC3B,WAAW,EAAE,gBAAgB;YAC7B,aAAa,EAAE,KAAK;YACpB,GAAG,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC;YACtE,MAAM,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC;SAClC,CAAC,CAAC;QACH,IAAI,MAAM,CAAC,SAAS;YAAE,OAAO;QAC7B,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;QAC7C,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,cAAc,KAAK,EAAE,CAAC,CAAC;IACtC,CAAC,CAAC;IAEF,KAAK,CAAC,eAAe,CAAC,OAAO,EAAE;QAC9B,WAAW,EAAE,gCAAgC;QAC7C,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,GAAG,EAAE,EAAE;YAC7B,IAAI,GAAG,CAAC,IAAI,KAAK,KAAK,EAAE,CAAC;gBACxB,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,yCAAyC,EAAE,SAAS,CAAC,CAAC;gBACpE,OAAO;YACR,CAAC;YACD,MAAM,MAAM,GAAG,MAAM,gBAAgB,CAAC,GAAG,CAAC,CAAC;YAC3C,IAAI,CAAC,MAAM;gBAAE,OAAO;YACpB,MAAM,WAAW,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,EAAE,EAAE;gBACnC,MAAM,WAAW,GAAG,KAAK,IAA2C,EAAE;oBACrE,OAAO,IAAI,EAAE,CAAC;wBACb,IAAI,CAAC;4BACJ,OAAO,MAAM,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;wBACvC,CAAC;wBAAC,OAAO,KAAK,EAAE,CAAC;4BAChB,IAAI,CAAC,MAAM,EAAE,CAAC,eAAe,CAAC,MAAM,CAAC,SAAS,EAAE,sBAAsB,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,OAAO,EAAE,CAAC;gCAC7F,OAAO,SAAS,CAAC;4BAClB,CAAC;wBACF,CAAC;oBACF,CAAC;gBACF,CAAC,CAAC;gBAEF,IAAI,OAAO,GAAG,MAAM,WAAW,EAAE,CAAC;gBAClC,IAAI,CAAC,OAAO;oBAAE,OAAO;gBACrB,OAAO,IAAI,EAAE,CAAC;oBACb,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;oBAC9D,IAAI,MAAM,CAAC,IAAI,KAAK,OAAO;wBAAE,OAAO;oBACpC,IAAI,WAAoB,CAAC;oBACzB,IAAI,CAAC;wBACJ,IAAI,MAAM,CAAC,IAAI,KAAK,UAAU;4BAAE,MAAM,aAAa,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC;6BAChE,IAAI,aAAa,CAAC,MAAM,CAAC,KAAK,CAAC;4BAAE,MAAM,WAAW,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;6BAClF,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU;4BAChD,MAAM,SAAS,CAAC,GAAG,EAAE,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;;4BACpD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,EAAE,OAAO,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;oBACrF,CAAC;oBAAC,OAAO,KAAK,EAAE,CAAC;wBAChB,WAAW,GAAG,KAAK,CAAC;oBACrB,CAAC;oBACD,MAAM,SAAS,GAAG,MAAM,WAAW,EAAE,CAAC;oBACtC,IAAI,CAAC,SAAS;wBAAE,OAAO;oBACvB,OAAO,GAAG,SAAS,CAAC;oBACpB,IAAI,WAAW,IAAI,CAAC,iBAAiB,CAAC,WAAW,CAAC,EAAE,CAAC;wBACpD,GAAG,CAAC,EAAE,CAAC,MAAM,CAAC,WAAW,YAAY,KAAK,CAAC,CAAC,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,CAAC,EAAE,OAAO,CAAC,CAAC;oBAClG,CAAC;gBACF,CAAC;YACF,CAAC,CAAC,CAAC;QACJ,CAAC;KACD,CAAC,CAAC;AACJ,CAAC","sourcesContent":["import type { ExtensionAPI, ExtensionCommandContext } from \"../../core/extensions/types.ts\";\nimport { formatBytes, LlamaClient, type LlamaModelInfo, normalizeLlamaServerUrl } from \"./client.ts\";\nimport { findHuggingFaceToken, HuggingFaceClient } from \"./huggingface.ts\";\nimport { createLlamaProvider, LLAMA_PROVIDER_ID } from \"./provider.ts\";\nimport { type LlamaUi, runWithProgress, showLlamaUi } from \"./ui.ts\";\n\nfunction modelIsLoaded(model: LlamaModelInfo): boolean {\n\treturn model.status.value === \"loaded\" || model.status.value === \"sleeping\";\n}\n\nfunction isConnectionError(error: unknown): boolean {\n\tif (!(error instanceof Error)) return false;\n\tconst message = `${error.name} ${error.message}`.toLowerCase();\n\treturn message.includes(\"fetch failed\") || message.includes(\"timeout\") || message.includes(\"network\");\n}\n\nfunction connectionErrorMessage(error: unknown): string {\n\tif (isConnectionError(error)) return \"Could not connect to the server.\";\n\treturn error instanceof Error ? error.message : String(error);\n}\n\nfunction parseHuggingFaceModel(value: string): { repository: string; quantization?: string } {\n\tconst colon = value.indexOf(\":\", value.indexOf(\"/\") + 1);\n\treturn colon < 0\n\t\t? { repository: value }\n\t\t: { repository: value.slice(0, colon), quantization: value.slice(colon + 1) };\n}\n\nasync function configuredClient(ctx: ExtensionCommandContext): Promise<LlamaClient | undefined> {\n\tconst result = await ctx.modelRegistry.getProviderAuth(LLAMA_PROVIDER_ID);\n\tif (!result) {\n\t\tctx.ui.notify(`Configure llama.cpp with /login ${LLAMA_PROVIDER_ID}`, \"warning\");\n\t\treturn undefined;\n\t}\n\tconst configuredUrl = result.env?.LLAMA_BASE_URL;\n\tconst serverUrl = normalizeLlamaServerUrl(\n\t\ttypeof configuredUrl === \"string\" && configuredUrl ? configuredUrl : (result.auth.baseUrl ?? \"\"),\n\t);\n\treturn new LlamaClient(serverUrl, result.auth.apiKey);\n}\n\nexport default function llamaExtension(relay: ExtensionAPI): void {\n\tconst provider = createLlamaProvider();\n\trelay.registerProvider(provider.provider);\n\n\tconst syncCatalog = async (\n\t\tctx: ExtensionCommandContext,\n\t\tclient: LlamaClient,\n\t\tcatalog?: LlamaModelInfo[],\n\t): Promise<LlamaModelInfo[]> => {\n\t\tconst signal = AbortSignal.timeout(15_000);\n\t\tconst current = catalog ?? (await client.list({ signal }));\n\t\tprovider.setCatalog(current, client.serverUrl);\n\t\tconst result = await ctx.modelRegistry.refresh({\n\t\t\tproviders: [LLAMA_PROVIDER_ID],\n\t\t\t// /llama already contacted the configured llama.cpp server, so keep this refresh live even in RELAY_OFFLINE.\n\t\t\tallowNetwork: true,\n\t\t\tsignal,\n\t\t});\n\t\tif (result.aborted) throw new Error(\"Model catalog refresh timed out.\");\n\t\tconst refreshError = result.errors.get(LLAMA_PROVIDER_ID);\n\t\tif (refreshError) throw refreshError;\n\t\treturn current;\n\t};\n\n\tconst loadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tcatalog: LlamaModelInfo[],\n\t\ttarget: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tconst loaded = catalog.filter((model) => model.id !== target.id && modelIsLoaded(model));\n\t\tlet replace = false;\n\t\tif (loaded.length > 0) {\n\t\t\tconst choice = await ui.select(`${loaded.length} model${loaded.length === 1 ? \" is\" : \"s are\"} loaded`, [\n\t\t\t\t\"Unload all and load\",\n\t\t\t\t\"Keep loaded and load\",\n\t\t\t\t\"Cancel\",\n\t\t\t]);\n\t\t\tif (!choice || choice === \"Cancel\") return;\n\t\t\treplace = choice === \"Unload all and load\";\n\t\t}\n\n\t\tconst restoreLoaded = async (): Promise<void> => {\n\t\t\tctx.ui.notify(\"Restoring previously loaded models\");\n\t\t\tfor (const model of loaded) await client.loadAndWait(model.id, () => {});\n\t\t\tawait syncCatalog(ctx, client);\n\t\t};\n\t\tif (replace) {\n\t\t\tfor (const model of loaded) await client.unloadAndWait(model.id);\n\t\t}\n\n\t\ttry {\n\t\t\tconst result = await runWithProgress(ui, {\n\t\t\t\ttitle: \"Loading model\",\n\t\t\t\tmodel: target.id,\n\t\t\t\tinitialMessage: \"Starting…\",\n\t\t\t\tcancelTitle: \"Stop loading?\",\n\t\t\t\tcancelMessage: target.id,\n\t\t\t\trun: (signal, update) => client.loadAndWait(target.id, update, signal),\n\t\t\t\tcancel: () => client.unload(target.id),\n\t\t\t});\n\t\t\tif (result.cancelled) {\n\t\t\t\tif (replace) await restoreLoaded();\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst refreshed = await syncCatalog(ctx, client);\n\t\t\tconst loadedModel = refreshed.find((model) => model.id === target.id);\n\t\t\tctx.ui.notify(\n\t\t\t\tloadedModel?.status.value === \"loaded\" ? `Loaded ${target.id}` : `Load started for ${target.id}`,\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tif (replace) {\n\t\t\t\ttry {\n\t\t\t\t\tawait restoreLoaded();\n\t\t\t\t} catch {\n\t\t\t\t\t// Preserve the original load error.\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t};\n\n\tconst unloadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tmodel: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tif (!(await ui.confirm(\"Unload model?\", model.id))) return;\n\t\tawait client.unloadAndWait(model.id);\n\t\tawait syncCatalog(ctx, client);\n\t\tctx.ui.notify(`Unloaded ${model.id}`);\n\t};\n\n\tconst downloadModel = async (ctx: ExtensionCommandContext, ui: LlamaUi, client: LlamaClient): Promise<void> => {\n\t\tconst huggingFace = new HuggingFaceClient(await findHuggingFaceToken());\n\t\tconst selected = await ui.searchModels((query, signal) => huggingFace.search(query, signal));\n\t\tif (!selected) return;\n\t\tconst parsed = parseHuggingFaceModel(selected);\n\t\tui.showStatus(\"Loading model details\", parsed.repository);\n\t\tconst details = await huggingFace.details(parsed.repository);\n\t\tif (details.gated) {\n\t\t\tconst approval = details.gated === \"manual\" ? \"Manual approval is required\" : \"Accept the access terms\";\n\t\t\tconst choice = await ui.select(\n\t\t\t\t`Hugging Face access required\\n${details.id}\\n\\n${approval} at:\\nhttps://huggingface.co/${details.id}\\n\\nThe llama.cpp server needs HF_TOKEN with access.`,\n\t\t\t\t[\"Continue\", \"Back\"],\n\t\t\t);\n\t\t\tif (choice !== \"Continue\") return;\n\t\t}\n\t\tlet quantization = parsed.quantization;\n\t\tif (!quantization && details.quantizations.length > 0) {\n\t\t\tconst options = details.quantizations.map((entry) => {\n\t\t\t\tconst detail = [\n\t\t\t\t\tentry.size === undefined ? undefined : formatBytes(entry.size),\n\t\t\t\t\tentry.name === \"Q4_K_M\" ? \"recommended\" : undefined,\n\t\t\t\t]\n\t\t\t\t\t.filter((value): value is string => Boolean(value))\n\t\t\t\t\t.join(\" · \");\n\t\t\t\treturn detail ? `${entry.name} · ${detail}` : entry.name;\n\t\t\t});\n\t\t\tconst choice = await ui.select(`Select quantization\\n${details.id}`, options);\n\t\t\tif (!choice) return;\n\t\t\tquantization = details.quantizations[options.indexOf(choice)]?.name;\n\t\t\tif (!quantization) return;\n\t\t}\n\t\tconst model = quantization ? `${details.id}:${quantization}` : details.id;\n\t\tconst result = await runWithProgress(ui, {\n\t\t\ttitle: \"Downloading model\",\n\t\t\tmodel,\n\t\t\tinitialMessage: \"Starting…\",\n\t\t\tcancelTitle: \"Stop download?\",\n\t\t\tcancelMessage: model,\n\t\t\trun: (signal, update) => client.downloadAndWait(model, update, signal),\n\t\t\tcancel: () => client.unload(model),\n\t\t});\n\t\tif (result.cancelled) return;\n\t\tawait syncCatalog(ctx, client, result.value);\n\t\tctx.ui.notify(`Downloaded ${model}`);\n\t};\n\n\trelay.registerCommand(\"llama\", {\n\t\tdescription: \"Manage llama.cpp router models\",\n\t\thandler: async (_args, ctx) => {\n\t\t\tif (ctx.mode !== \"tui\") {\n\t\t\t\tctx.ui.notify(\"/llama is available in interactive mode\", \"warning\");\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst client = await configuredClient(ctx);\n\t\t\tif (!client) return;\n\t\t\tawait showLlamaUi(ctx, async (ui) => {\n\t\t\t\tconst readCatalog = async (): Promise<LlamaModelInfo[] | undefined> => {\n\t\t\t\t\twhile (true) {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\treturn await syncCatalog(ctx, client);\n\t\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\t\tif ((await ui.connectionError(client.serverUrl, connectionErrorMessage(error))) === \"close\") {\n\t\t\t\t\t\t\t\treturn undefined;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t};\n\n\t\t\t\tlet catalog = await readCatalog();\n\t\t\t\tif (!catalog) return;\n\t\t\t\twhile (true) {\n\t\t\t\t\tconst action = await ui.showModels(client.serverUrl, catalog);\n\t\t\t\t\tif (action.type === \"close\") return;\n\t\t\t\t\tlet actionError: unknown;\n\t\t\t\t\ttry {\n\t\t\t\t\t\tif (action.type === \"download\") await downloadModel(ctx, ui, client);\n\t\t\t\t\t\telse if (modelIsLoaded(action.model)) await unloadModel(ctx, ui, client, action.model);\n\t\t\t\t\t\telse if (action.model.status.value === \"unloaded\")\n\t\t\t\t\t\t\tawait loadModel(ctx, ui, client, catalog, action.model);\n\t\t\t\t\t\telse ctx.ui.notify(`${action.model.id} is ${action.model.status.value}`, \"warning\");\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tactionError = error;\n\t\t\t\t\t}\n\t\t\t\t\tconst refreshed = await readCatalog();\n\t\t\t\t\tif (!refreshed) return;\n\t\t\t\t\tcatalog = refreshed;\n\t\t\t\t\tif (actionError && !isConnectionError(actionError)) {\n\t\t\t\t\t\tctx.ui.notify(actionError instanceof Error ? actionError.message : String(actionError), \"error\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t});\n\t\t},\n\t});\n}\n"]}

package/dist/extensions/llama/provider.d.ts

@@ -0,0 +1,12 @@
+import { type Provider } from "@relay-harness/ai";
+import { type LlamaModelInfo } from "./client.ts";
+export declare const LLAMA_PROVIDER_ID = "llama.cpp";
+export declare const DEFAULT_LLAMA_SERVER_URL = "http://127.0.0.1:8080";
+export interface LlamaProviderController {
+    provider: Provider<"openai-completions">;
+    setCatalog(models: readonly LlamaModelInfo[], serverUrl: string, options?: {
+        routerAutoload?: boolean;
+    }): void;
+}
+export declare function createLlamaProvider(): LlamaProviderController;
+//# sourceMappingURL=provider.d.ts.map

package/dist/extensions/llama/provider.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"provider.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/provider.ts"],"names":[],"mappings":"AAAA,OAAO,EAON,KAAK,QAAQ,EAGb,MAAM,mBAAmB,CAAC;AAG3B,OAAO,EAEN,KAAK,cAAc,EAInB,MAAM,aAAa,CAAC;AAErB,eAAO,MAAM,iBAAiB,cAAc,CAAC;AAC7C,eAAO,MAAM,wBAAwB,0BAA0B,CAAC;AA6GhE,MAAM,WAAW,uBAAuB;IACvC,QAAQ,EAAE,QAAQ,CAAC,oBAAoB,CAAC,CAAC;IACzC,UAAU,CAAC,MAAM,EAAE,SAAS,cAAc,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,cAAc,CAAC,EAAE,OAAO,CAAA;KAAE,GAAG,IAAI,CAAC;CAC/G;AAED,wBAAgB,mBAAmB,IAAI,uBAAuB,CAkI7D"}

package/dist/extensions/llama/provider.js

@@ -0,0 +1,213 @@
+import { isModelType, } from "@relay-harness/ai";
+import { llamaCppClassifyApi } from "@relay-harness/ai/api/llama-cpp-classify.lazy";
+import { stream, streamSimple } from "@relay-harness/ai/compat";
+import { LlamaClient, llamaInferenceUrl, normalizeLlamaServerUrl, } from "./client.js";
+export const LLAMA_PROVIDER_ID = "llama.cpp";
+export const DEFAULT_LLAMA_SERVER_URL = "http://127.0.0.1:8080";
+function credentialServerUrl(credential) {
+    const value = credential?.env?.LLAMA_BASE_URL;
+    return typeof value === "string" && value.trim() ? normalizeLlamaServerUrl(value) : undefined;
+}
+async function resolveServerUrl(ctx, credential) {
+    const configured = credentialServerUrl(credential) ?? (await ctx.env("LLAMA_BASE_URL"))?.trim();
+    return configured ? normalizeLlamaServerUrl(configured) : undefined;
+}
+function modelIsSelectable(model, routerAutoload) {
+    if (model.status.value === "loaded")
+        return true;
+    // llama.cpp reports idle-slept models as "sleeping"; requests wake them automatically.
+    if (model.status.value === "sleeping")
+        return true;
+    // Unloaded presets are routable only when llama.cpp router autoload can load them on first use.
+    return routerAutoload && model.status.value === "unloaded" && !model.status.failed && model.source === "preset";
+}
+async function routerAutoloadEnabled(client, catalog, signal) {
+    if (!catalog.some((model) => model.status.value === "unloaded" && model.source === "preset"))
+        return false;
+    try {
+        return (await client.props({ signal })).models_autoload === true;
+    }
+    catch {
+        return false;
+    }
+}
+function configuredContextWindow(model) {
+    const args = model.status.args ?? [];
+    for (let index = 0; index < args.length - 1; index++) {
+        const flag = args[index];
+        if (flag !== "--ctx-size" && flag !== "-c" && flag !== "-ctx")
+            continue;
+        const contextWindow = Number(args[index + 1]);
+        if (Number.isSafeInteger(contextWindow) && contextWindow > 0)
+            return contextWindow;
+    }
+    return undefined;
+}
+function contextWindowOf(model, cachedContextWindow) {
+    const runtimeContextWindow = model.meta?.n_ctx;
+    if (runtimeContextWindow && runtimeContextWindow > 0)
+        return runtimeContextWindow;
+    const configuredContext = configuredContextWindow(model);
+    if (configuredContext)
+        return configuredContext;
+    if (cachedContextWindow && cachedContextWindow > 0)
+        return cachedContextWindow;
+    const trainingContextWindow = model.meta?.n_ctx_train;
+    return trainingContextWindow && trainingContextWindow > 0 ? trainingContextWindow : 128000;
+}
+/** The same llama.cpp model used as a classifier: answers are read from next-token label probabilities. */
+function toRelayClassifierModel(model, serverUrl, cachedContextWindow) {
+    return {
+        type: "classifier",
+        id: model.id,
+        name: model.id,
+        api: "llama-cpp-classify",
+        provider: LLAMA_PROVIDER_ID,
+        baseUrl: serverUrl,
+        input: ["text"],
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+        contextWindow: contextWindowOf(model, cachedContextWindow),
+    };
+}
+function toRelayModel(model, serverUrl, props, cachedContextWindow) {
+    const contextWindow = contextWindowOf(model, cachedContextWindow);
+    const reasoning = props?.chat_template?.includes("enable_thinking") === true;
+    return {
+        id: model.id,
+        name: model.id,
+        api: "openai-completions",
+        provider: LLAMA_PROVIDER_ID,
+        baseUrl: llamaInferenceUrl(serverUrl),
+        reasoning,
+        ...(reasoning && {
+            thinkingLevelMap: { off: "off", minimal: null, low: null, medium: "medium", high: null, xhigh: null },
+        }),
+        input: model.architecture?.input_modalities?.includes("image") ? ["text", "image"] : ["text"],
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+        contextWindow,
+        maxTokens: contextWindow,
+        compat: {
+            supportsStore: false,
+            supportsDeveloperRole: false,
+            supportsReasoningEffort: false,
+            supportsUsageInStreaming: true,
+            supportsStrictMode: false,
+            maxTokensField: "max_tokens",
+            ...(reasoning && { thinkingFormat: "qwen-chat-template" }),
+        },
+    };
+}
+export function createLlamaProvider() {
+    let models = [];
+    let classifiers = [];
+    const classifier = llamaCppClassifyApi();
+    const setCatalog = (catalog, serverUrl, options = {}) => {
+        const selectable = catalog.filter((model) => modelIsSelectable(model, options.routerAutoload === true));
+        models = selectable.map((model) => toRelayModel(model, serverUrl));
+        classifiers = selectable.map((model) => toRelayClassifierModel(model, serverUrl));
+    };
+    const provider = {
+        id: LLAMA_PROVIDER_ID,
+        name: "llama.cpp",
+        baseUrl: llamaInferenceUrl(DEFAULT_LLAMA_SERVER_URL),
+        auth: {
+            apiKey: {
+                name: "llama.cpp server",
+                login: async (interaction) => {
+                    const enteredUrl = await interaction.prompt({
+                        type: "text",
+                        message: "llama.cpp server URL",
+                        placeholder: process.env.LLAMA_BASE_URL ?? DEFAULT_LLAMA_SERVER_URL,
+                    });
+                    const serverUrl = normalizeLlamaServerUrl(enteredUrl.trim() || process.env.LLAMA_BASE_URL || DEFAULT_LLAMA_SERVER_URL);
+                    const apiKey = (await interaction.prompt({
+                        type: "secret",
+                        message: "API key (optional)",
+                    })).trim();
+                    await new LlamaClient(serverUrl, apiKey || undefined).list({ signal: interaction.signal });
+                    return {
+                        type: "api_key",
+                        key: apiKey || undefined,
+                        env: { LLAMA_BASE_URL: serverUrl },
+                    };
+                },
+                check: async ({ ctx, credential }) => {
+                    const serverUrl = await resolveServerUrl(ctx, credential);
+                    return serverUrl
+                        ? { type: "api_key", source: credential ? "stored credential" : "LLAMA_BASE_URL" }
+                        : undefined;
+                },
+                resolve: async ({ ctx, credential }) => {
+                    const serverUrl = await resolveServerUrl(ctx, credential);
+                    if (!serverUrl)
+                        return undefined;
+                    const apiKey = credential?.key ?? (await ctx.env("LLAMA_API_KEY")) ?? "local";
+                    return {
+                        auth: { apiKey, baseUrl: llamaInferenceUrl(serverUrl) },
+                        env: { ...credential?.env, LLAMA_BASE_URL: serverUrl },
+                        source: credential ? "stored credential" : "LLAMA_BASE_URL",
+                    };
+                },
+            },
+        },
+        getModels: () => models,
+        getAllModels: () => [...models, ...classifiers],
+        refreshModels: async (context) => {
+            const cachedContextWindows = new Map();
+            if (context.stored) {
+                const stored = context.stored.models.filter((model) => model.provider === LLAMA_PROVIDER_ID);
+                const restored = stored.filter((model) => isModelType(model, "chat") && model.api === "openai-completions");
+                const restoredClassifiers = stored.filter((model) => isModelType(model, "classifier") && model.api === "llama-cpp-classify");
+                for (const model of [...restored, ...restoredClassifiers]) {
+                    cachedContextWindows.set(model.id, model.contextWindow);
+                }
+                if (!(await context.publish({
+                    update: () => {
+                        models = restored;
+                        classifiers = restoredClassifiers;
+                    },
+                }))) {
+                    return;
+                }
+            }
+            if (!context.allowNetwork || context.signal.aborted || context.credential?.type !== "api_key")
+                return;
+            const serverUrl = credentialServerUrl(context.credential);
+            if (!serverUrl)
+                return;
+            const client = new LlamaClient(serverUrl, context.credential.key);
+            const catalog = await client.list({ signal: context.signal });
+            if (context.signal.aborted)
+                return;
+            const routerAutoload = await routerAutoloadEnabled(client, catalog, context.signal);
+            if (context.signal.aborted)
+                return;
+            const selectable = catalog.filter((model) => modelIsSelectable(model, routerAutoload));
+            const refreshed = await Promise.all(selectable.map(async (model) => {
+                const cachedContextWindow = cachedContextWindows.get(model.id);
+                // Only loaded models expose their template without side effects. Unloaded autoload presets
+                // would need to be loaded, while querying sleeping models may wake them. Those models remain
+                // unclassified until they are loaded or woken and a later catalog refresh discovers them.
+                if (model.status.value !== "loaded")
+                    return toRelayModel(model, serverUrl, undefined, cachedContextWindow);
+                const props = await client.props({ model: model.id, signal: context.signal });
+                return toRelayModel(model, serverUrl, props, cachedContextWindow);
+            }));
+            const refreshedClassifiers = selectable.map((model) => toRelayClassifierModel(model, serverUrl, cachedContextWindows.get(model.id)));
+            if (context.signal.aborted)
+                return;
+            await context.publish({
+                persist: { models: [...refreshed, ...refreshedClassifiers], checkedAt: Date.now() },
+                update: () => {
+                    models = refreshed;
+                    classifiers = refreshedClassifiers;
+                },
+            });
+        },
+        stream: (model, context, options) => stream(model, context, options),
+        streamSimple: (model, context, options) => streamSimple(model, context, options),
+        classify: (model, context, options) => classifier.classify(model, context, options),
+    };
+    return { provider, setCatalog };
+}
+//# sourceMappingURL=provider.js.map

package/dist/extensions/llama/provider.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"provider.js","sourceRoot":"","sources":["../../../src/extensions/llama/provider.ts"],"names":[],"mappings":"AAAA,OAAO,EAKN,WAAW,GAKX,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EAAE,mBAAmB,EAAE,MAAM,+CAA+C,CAAC;AACpF,OAAO,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAChE,OAAO,EACN,WAAW,EAGX,iBAAiB,EACjB,uBAAuB,GACvB,MAAM,aAAa,CAAC;AAErB,MAAM,CAAC,MAAM,iBAAiB,GAAG,WAAW,CAAC;AAC7C,MAAM,CAAC,MAAM,wBAAwB,GAAG,uBAAuB,CAAC;AAChE,SAAS,mBAAmB,CAAC,UAAwC;IACpE,MAAM,KAAK,GAAG,UAAU,EAAE,GAAG,EAAE,cAAc,CAAC;IAC9C,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,uBAAuB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAC/F,CAAC;AAED,KAAK,UAAU,gBAAgB,CAC9B,GAAgB,EAChB,UAAwC;IAExC,MAAM,UAAU,GAAG,mBAAmB,CAAC,UAAU,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC;IAChG,OAAO,UAAU,CAAC,CAAC,CAAC,uBAAuB,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AACrE,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAqB,EAAE,cAAuB;IACxE,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ;QAAE,OAAO,IAAI,CAAC;IACjD,uFAAuF;IACvF,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU;QAAE,OAAO,IAAI,CAAC;IACnD,gGAAgG;IAChG,OAAO,cAAc,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,IAAI,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;AACjH,CAAC;AAED,KAAK,UAAU,qBAAqB,CACnC,MAAmB,EACnB,OAAkC,EAClC,MAAmB;IAEnB,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU,IAAI,KAAK,CAAC,MAAM,KAAK,QAAQ,CAAC;QAAE,OAAO,KAAK,CAAC;IAC3G,IAAI,CAAC;QACJ,OAAO,CAAC,MAAM,MAAM,CAAC,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,eAAe,KAAK,IAAI,CAAC;IAClE,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,KAAK,CAAC;IACd,CAAC;AACF,CAAC;AAED,SAAS,uBAAuB,CAAC,KAAqB;IACrD,MAAM,IAAI,GAAG,KAAK,CAAC,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC;IACrC,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,KAAK,EAAE,EAAE,CAAC;QACtD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;QACzB,IAAI,IAAI,KAAK,YAAY,IAAI,IAAI,KAAK,IAAI,IAAI,IAAI,KAAK,MAAM;YAAE,SAAS;QACxE,MAAM,aAAa,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC;QAC9C,IAAI,MAAM,CAAC,aAAa,CAAC,aAAa,CAAC,IAAI,aAAa,GAAG,CAAC;YAAE,OAAO,aAAa,CAAC;IACpF,CAAC;IACD,OAAO,SAAS,CAAC;AAClB,CAAC;AAED,SAAS,eAAe,CAAC,KAAqB,EAAE,mBAA4B;IAC3E,MAAM,oBAAoB,GAAG,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC;IAC/C,IAAI,oBAAoB,IAAI,oBAAoB,GAAG,CAAC;QAAE,OAAO,oBAAoB,CAAC;IAClF,MAAM,iBAAiB,GAAG,uBAAuB,CAAC,KAAK,CAAC,CAAC;IACzD,IAAI,iBAAiB;QAAE,OAAO,iBAAiB,CAAC;IAChD,IAAI,mBAAmB,IAAI,mBAAmB,GAAG,CAAC;QAAE,OAAO,mBAAmB,CAAC;IAC/E,MAAM,qBAAqB,GAAG,KAAK,CAAC,IAAI,EAAE,WAAW,CAAC;IACtD,OAAO,qBAAqB,IAAI,qBAAqB,GAAG,CAAC,CAAC,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC,MAAM,CAAC;AAC5F,CAAC;AAED,2GAA2G;AAC3G,SAAS,sBAAsB,CAC9B,KAAqB,EACrB,SAAiB,EACjB,mBAA4B;IAE5B,OAAO;QACN,IAAI,EAAE,YAAY;QAClB,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,IAAI,EAAE,KAAK,CAAC,EAAE;QACd,GAAG,EAAE,oBAAoB;QACzB,QAAQ,EAAE,iBAAiB;QAC3B,OAAO,EAAE,SAAS;QAClB,KAAK,EAAE,CAAC,MAAM,CAAC;QACf,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE;QAC1D,aAAa,EAAE,eAAe,CAAC,KAAK,EAAE,mBAAmB,CAAC;KAC1D,CAAC;AACH,CAAC;AAED,SAAS,YAAY,CACpB,KAAqB,EACrB,SAAiB,EACjB,KAAwB,EACxB,mBAA4B;IAE5B,MAAM,aAAa,GAAG,eAAe,CAAC,KAAK,EAAE,mBAAmB,CAAC,CAAC;IAClE,MAAM,SAAS,GAAG,KAAK,EAAE,aAAa,EAAE,QAAQ,CAAC,iBAAiB,CAAC,KAAK,IAAI,CAAC;IAC7E,OAAO;QACN,EAAE,EAAE,KAAK,CAAC,EAAE;QACZ,IAAI,EAAE,KAAK,CAAC,EAAE;QACd,GAAG,EAAE,oBAAoB;QACzB,QAAQ,EAAE,iBAAiB;QAC3B,OAAO,EAAE,iBAAiB,CAAC,SAAS,CAAC;QACrC,SAAS;QACT,GAAG,CAAC,SAAS,IAAI;YAChB,gBAAgB,EAAE,EAAE,GAAG,EAAE,KAAK,EAAE,OAAO,EAAE,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE;SACrG,CAAC;QACF,KAAK,EAAE,KAAK,CAAC,YAAY,EAAE,gBAAgB,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC;QAC7F,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE;QAC1D,aAAa;QACb,SAAS,EAAE,aAAa;QACxB,MAAM,EAAE;YACP,aAAa,EAAE,KAAK;YACpB,qBAAqB,EAAE,KAAK;YAC5B,uBAAuB,EAAE,KAAK;YAC9B,wBAAwB,EAAE,IAAI;YAC9B,kBAAkB,EAAE,KAAK;YACzB,cAAc,EAAE,YAAY;YAC5B,GAAG,CAAC,SAAS,IAAI,EAAE,cAAc,EAAE,oBAAoB,EAAE,CAAC;SAC1D;KACD,CAAC;AACH,CAAC;AAOD,MAAM,UAAU,mBAAmB;IAClC,IAAI,MAAM,GAA2C,EAAE,CAAC;IACxD,IAAI,WAAW,GAAqD,EAAE,CAAC;IACvE,MAAM,UAAU,GAAG,mBAAmB,EAAE,CAAC;IAEzC,MAAM,UAAU,GAAG,CAClB,OAAkC,EAClC,SAAiB,EACjB,OAAO,GAAiC,EAAE,EACnC,EAAE;QACT,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,iBAAiB,CAAC,KAAK,EAAE,OAAO,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC,CAAC;QACxG,MAAM,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,YAAY,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC,CAAC;QACnE,WAAW,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,sBAAsB,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC,CAAC;IACnF,CAAC,CAAC;IAEF,MAAM,QAAQ,GAAmC;QAChD,EAAE,EAAE,iBAAiB;QACrB,IAAI,EAAE,WAAW;QACjB,OAAO,EAAE,iBAAiB,CAAC,wBAAwB,CAAC;QACpD,IAAI,EAAE;YACL,MAAM,EAAE;gBACP,IAAI,EAAE,kBAAkB;gBACxB,KAAK,EAAE,KAAK,EAAE,WAAW,EAA6B,EAAE;oBACvD,MAAM,UAAU,GAAG,MAAM,WAAW,CAAC,MAAM,CAAC;wBAC3C,IAAI,EAAE,MAAM;wBACZ,OAAO,EAAE,sBAAsB;wBAC/B,WAAW,EAAE,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,wBAAwB;qBACnE,CAAC,CAAC;oBACH,MAAM,SAAS,GAAG,uBAAuB,CACxC,UAAU,CAAC,IAAI,EAAE,IAAI,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,wBAAwB,CAC3E,CAAC;oBACF,MAAM,MAAM,GAAG,CACd,MAAM,WAAW,CAAC,MAAM,CAAC;wBACxB,IAAI,EAAE,QAAQ;wBACd,OAAO,EAAE,oBAAoB;qBAC7B,CAAC,CACF,CAAC,IAAI,EAAE,CAAC;oBACT,MAAM,IAAI,WAAW,CAAC,SAAS,EAAE,MAAM,IAAI,SAAS,CAAC,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,WAAW,CAAC,MAAM,EAAE,CAAC,CAAC;oBAC3F,OAAO;wBACN,IAAI,EAAE,SAAS;wBACf,GAAG,EAAE,MAAM,IAAI,SAAS;wBACxB,GAAG,EAAE,EAAE,cAAc,EAAE,SAAS,EAAE;qBAClC,CAAC;gBACH,CAAC;gBACD,KAAK,EAAE,KAAK,EAAE,EAAE,GAAG,EAAE,UAAU,EAAE,EAAE,EAAE;oBACpC,MAAM,SAAS,GAAG,MAAM,gBAAgB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;oBAC1D,OAAO,SAAS;wBACf,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,gBAAgB,EAAE;wBAClF,CAAC,CAAC,SAAS,CAAC;gBACd,CAAC;gBACD,OAAO,EAAE,KAAK,EAAE,EAAE,GAAG,EAAE,UAAU,EAAE,EAAmC,EAAE;oBACvE,MAAM,SAAS,GAAG,MAAM,gBAAgB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;oBAC1D,IAAI,CAAC,SAAS;wBAAE,OAAO,SAAS,CAAC;oBACjC,MAAM,MAAM,GAAG,UAAU,EAAE,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC,IAAI,OAAO,CAAC;oBAC9E,OAAO;wBACN,IAAI,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,iBAAiB,CAAC,SAAS,CAAC,EAAE;wBACvD,GAAG,EAAE,EAAE,GAAG,UAAU,EAAE,GAAG,EAAE,cAAc,EAAE,SAAS,EAAE;wBACtD,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,gBAAgB;qBAC3D,CAAC;gBACH,CAAC;aACD;SACD;QACD,SAAS,EAAE,GAAG,EAAE,CAAC,MAAM;QACvB,YAAY,EAAE,GAAG,EAAE,CAAC,CAAC,GAAG,MAAM,EAAE,GAAG,WAAW,CAAC;QAC/C,aAAa,EAAE,KAAK,EAAE,OAA6B,EAAiB,EAAE;YACrE,MAAM,oBAAoB,GAAG,IAAI,GAAG,EAAkB,CAAC;YACvD,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;gBACpB,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,iBAAiB,CAAC,CAAC;gBAC7F,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAC7B,CAAC,KAAK,EAAwC,EAAE,CAC/C,WAAW,CAAC,KAAK,EAAE,MAAM,CAAC,IAAI,KAAK,CAAC,GAAG,KAAK,oBAAoB,CACjE,CAAC;gBACF,MAAM,mBAAmB,GAAG,MAAM,CAAC,MAAM,CACxC,CAAC,KAAK,EAAkD,EAAE,CACzD,WAAW,CAAC,KAAK,EAAE,YAAY,CAAC,IAAI,KAAK,CAAC,GAAG,KAAK,oBAAoB,CACvE,CAAC;gBACF,KAAK,MAAM,KAAK,IAAI,CAAC,GAAG,QAAQ,EAAE,GAAG,mBAAmB,CAAC,EAAE,CAAC;oBAC3D,oBAAoB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,aAAa,CAAC,CAAC;gBACzD,CAAC;gBACD,IACC,CAAC,CAAC,MAAM,OAAO,CAAC,OAAO,CAAC;oBACvB,MAAM,EAAE,GAAG,EAAE;wBACZ,MAAM,GAAG,QAAQ,CAAC;wBAClB,WAAW,GAAG,mBAAmB,CAAC;oBACnC,CAAC;iBACD,CAAC,CAAC,EACF,CAAC;oBACF,OAAO;gBACR,CAAC;YACF,CAAC;YAED,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,IAAI,OAAO,CAAC,UAAU,EAAE,IAAI,KAAK,SAAS;gBAAE,OAAO;YACtG,MAAM,SAAS,GAAG,mBAAmB,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;YAC1D,IAAI,CAAC,SAAS;gBAAE,OAAO;YACvB,MAAM,MAAM,GAAG,IAAI,WAAW,CAAC,SAAS,EAAE,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC;YAClE,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;YAC9D,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,cAAc,GAAG,MAAM,qBAAqB,CAAC,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;YACpF,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,iBAAiB,CAAC,KAAK,EAAE,cAAc,CAAC,CAAC,CAAC;YACvF,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,GAAG,CAClC,UAAU,CAAC,GAAG,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE;gBAC9B,MAAM,mBAAmB,GAAG,oBAAoB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;gBAC/D,2FAA2F;gBAC3F,6FAA6F;gBAC7F,0FAA0F;gBAC1F,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,QAAQ;oBAClC,OAAO,YAAY,CAAC,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,mBAAmB,CAAC,CAAC;gBACvE,MAAM,KAAK,GAAG,MAAM,MAAM,CAAC,KAAK,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;gBAC9E,OAAO,YAAY,CAAC,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,mBAAmB,CAAC,CAAC;YACnE,CAAC,CAAC,CACF,CAAC;YACF,MAAM,oBAAoB,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CACrD,sBAAsB,CAAC,KAAK,EAAE,SAAS,EAAE,oBAAoB,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAC5E,CAAC;YACF,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO;gBAAE,OAAO;YACnC,MAAM,OAAO,CAAC,OAAO,CAAC;gBACrB,OAAO,EAAE,EAAE,MAAM,EAAE,CAAC,GAAG,SAAS,EAAE,GAAG,oBAAoB,CAAC,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE;gBACnF,MAAM,EAAE,GAAG,EAAE;oBACZ,MAAM,GAAG,SAAS,CAAC;oBACnB,WAAW,GAAG,oBAAoB,CAAC;gBACpC,CAAC;aACD,CAAC,CAAC;QACJ,CAAC;QACD,MAAM,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,OAA4C,CAAC;QACzG,YAAY,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC;QAChF,QAAQ,EAAE,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC;KACnF,CAAC;IAEF,OAAO,EAAE,QAAQ,EAAE,UAAU,EAAE,CAAC;AACjC,CAAC","sourcesContent":["import {\n\ttype ApiKeyCredential,\n\ttype AuthContext,\n\ttype AuthResult,\n\ttype ClassifierModel,\n\tisModelType,\n\ttype Model,\n\ttype Provider,\n\ttype ProviderStreamOptions,\n\ttype RefreshModelsContext,\n} from \"@relay-harness/ai\";\nimport { llamaCppClassifyApi } from \"@relay-harness/ai/api/llama-cpp-classify.lazy\";\nimport { stream, streamSimple } from \"@relay-harness/ai/compat\";\nimport {\n\tLlamaClient,\n\ttype LlamaModelInfo,\n\ttype LlamaServerProps,\n\tllamaInferenceUrl,\n\tnormalizeLlamaServerUrl,\n} from \"./client.ts\";\n\nexport const LLAMA_PROVIDER_ID = \"llama.cpp\";\nexport const DEFAULT_LLAMA_SERVER_URL = \"http://127.0.0.1:8080\";\nfunction credentialServerUrl(credential: ApiKeyCredential | undefined): string | undefined {\n\tconst value = credential?.env?.LLAMA_BASE_URL;\n\treturn typeof value === \"string\" && value.trim() ? normalizeLlamaServerUrl(value) : undefined;\n}\n\nasync function resolveServerUrl(\n\tctx: AuthContext,\n\tcredential: ApiKeyCredential | undefined,\n): Promise<string | undefined> {\n\tconst configured = credentialServerUrl(credential) ?? (await ctx.env(\"LLAMA_BASE_URL\"))?.trim();\n\treturn configured ? normalizeLlamaServerUrl(configured) : undefined;\n}\n\nfunction modelIsSelectable(model: LlamaModelInfo, routerAutoload: boolean): boolean {\n\tif (model.status.value === \"loaded\") return true;\n\t// llama.cpp reports idle-slept models as \"sleeping\"; requests wake them automatically.\n\tif (model.status.value === \"sleeping\") return true;\n\t// Unloaded presets are routable only when llama.cpp router autoload can load them on first use.\n\treturn routerAutoload && model.status.value === \"unloaded\" && !model.status.failed && model.source === \"preset\";\n}\n\nasync function routerAutoloadEnabled(\n\tclient: LlamaClient,\n\tcatalog: readonly LlamaModelInfo[],\n\tsignal: AbortSignal,\n): Promise<boolean> {\n\tif (!catalog.some((model) => model.status.value === \"unloaded\" && model.source === \"preset\")) return false;\n\ttry {\n\t\treturn (await client.props({ signal })).models_autoload === true;\n\t} catch {\n\t\treturn false;\n\t}\n}\n\nfunction configuredContextWindow(model: LlamaModelInfo): number | undefined {\n\tconst args = model.status.args ?? [];\n\tfor (let index = 0; index < args.length - 1; index++) {\n\t\tconst flag = args[index];\n\t\tif (flag !== \"--ctx-size\" && flag !== \"-c\" && flag !== \"-ctx\") continue;\n\t\tconst contextWindow = Number(args[index + 1]);\n\t\tif (Number.isSafeInteger(contextWindow) && contextWindow > 0) return contextWindow;\n\t}\n\treturn undefined;\n}\n\nfunction contextWindowOf(model: LlamaModelInfo, cachedContextWindow?: number): number {\n\tconst runtimeContextWindow = model.meta?.n_ctx;\n\tif (runtimeContextWindow && runtimeContextWindow > 0) return runtimeContextWindow;\n\tconst configuredContext = configuredContextWindow(model);\n\tif (configuredContext) return configuredContext;\n\tif (cachedContextWindow && cachedContextWindow > 0) return cachedContextWindow;\n\tconst trainingContextWindow = model.meta?.n_ctx_train;\n\treturn trainingContextWindow && trainingContextWindow > 0 ? trainingContextWindow : 128000;\n}\n\n/** The same llama.cpp model used as a classifier: answers are read from next-token label probabilities. */\nfunction toRelayClassifierModel(\n\tmodel: LlamaModelInfo,\n\tserverUrl: string,\n\tcachedContextWindow?: number,\n): ClassifierModel<\"llama-cpp-classify\"> {\n\treturn {\n\t\ttype: \"classifier\",\n\t\tid: model.id,\n\t\tname: model.id,\n\t\tapi: \"llama-cpp-classify\",\n\t\tprovider: LLAMA_PROVIDER_ID,\n\t\tbaseUrl: serverUrl,\n\t\tinput: [\"text\"],\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },\n\t\tcontextWindow: contextWindowOf(model, cachedContextWindow),\n\t};\n}\n\nfunction toRelayModel(\n\tmodel: LlamaModelInfo,\n\tserverUrl: string,\n\tprops?: LlamaServerProps,\n\tcachedContextWindow?: number,\n): Model<\"openai-completions\"> {\n\tconst contextWindow = contextWindowOf(model, cachedContextWindow);\n\tconst reasoning = props?.chat_template?.includes(\"enable_thinking\") === true;\n\treturn {\n\t\tid: model.id,\n\t\tname: model.id,\n\t\tapi: \"openai-completions\",\n\t\tprovider: LLAMA_PROVIDER_ID,\n\t\tbaseUrl: llamaInferenceUrl(serverUrl),\n\t\treasoning,\n\t\t...(reasoning && {\n\t\t\tthinkingLevelMap: { off: \"off\", minimal: null, low: null, medium: \"medium\", high: null, xhigh: null },\n\t\t}),\n\t\tinput: model.architecture?.input_modalities?.includes(\"image\") ? [\"text\", \"image\"] : [\"text\"],\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },\n\t\tcontextWindow,\n\t\tmaxTokens: contextWindow,\n\t\tcompat: {\n\t\t\tsupportsStore: false,\n\t\t\tsupportsDeveloperRole: false,\n\t\t\tsupportsReasoningEffort: false,\n\t\t\tsupportsUsageInStreaming: true,\n\t\t\tsupportsStrictMode: false,\n\t\t\tmaxTokensField: \"max_tokens\",\n\t\t\t...(reasoning && { thinkingFormat: \"qwen-chat-template\" }),\n\t\t},\n\t};\n}\n\nexport interface LlamaProviderController {\n\tprovider: Provider<\"openai-completions\">;\n\tsetCatalog(models: readonly LlamaModelInfo[], serverUrl: string, options?: { routerAutoload?: boolean }): void;\n}\n\nexport function createLlamaProvider(): LlamaProviderController {\n\tlet models: readonly Model<\"openai-completions\">[] = [];\n\tlet classifiers: readonly ClassifierModel<\"llama-cpp-classify\">[] = [];\n\tconst classifier = llamaCppClassifyApi();\n\n\tconst setCatalog = (\n\t\tcatalog: readonly LlamaModelInfo[],\n\t\tserverUrl: string,\n\t\toptions: { routerAutoload?: boolean } = {},\n\t): void => {\n\t\tconst selectable = catalog.filter((model) => modelIsSelectable(model, options.routerAutoload === true));\n\t\tmodels = selectable.map((model) => toRelayModel(model, serverUrl));\n\t\tclassifiers = selectable.map((model) => toRelayClassifierModel(model, serverUrl));\n\t};\n\n\tconst provider: Provider<\"openai-completions\"> = {\n\t\tid: LLAMA_PROVIDER_ID,\n\t\tname: \"llama.cpp\",\n\t\tbaseUrl: llamaInferenceUrl(DEFAULT_LLAMA_SERVER_URL),\n\t\tauth: {\n\t\t\tapiKey: {\n\t\t\t\tname: \"llama.cpp server\",\n\t\t\t\tlogin: async (interaction): Promise<ApiKeyCredential> => {\n\t\t\t\t\tconst enteredUrl = await interaction.prompt({\n\t\t\t\t\t\ttype: \"text\",\n\t\t\t\t\t\tmessage: \"llama.cpp server URL\",\n\t\t\t\t\t\tplaceholder: process.env.LLAMA_BASE_URL ?? DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t});\n\t\t\t\t\tconst serverUrl = normalizeLlamaServerUrl(\n\t\t\t\t\t\tenteredUrl.trim() || process.env.LLAMA_BASE_URL || DEFAULT_LLAMA_SERVER_URL,\n\t\t\t\t\t);\n\t\t\t\t\tconst apiKey = (\n\t\t\t\t\t\tawait interaction.prompt({\n\t\t\t\t\t\t\ttype: \"secret\",\n\t\t\t\t\t\t\tmessage: \"API key (optional)\",\n\t\t\t\t\t\t})\n\t\t\t\t\t).trim();\n\t\t\t\t\tawait new LlamaClient(serverUrl, apiKey || undefined).list({ signal: interaction.signal });\n\t\t\t\t\treturn {\n\t\t\t\t\t\ttype: \"api_key\",\n\t\t\t\t\t\tkey: apiKey || undefined,\n\t\t\t\t\t\tenv: { LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t\tcheck: async ({ ctx, credential }) => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\treturn serverUrl\n\t\t\t\t\t\t? { type: \"api_key\", source: credential ? \"stored credential\" : \"LLAMA_BASE_URL\" }\n\t\t\t\t\t\t: undefined;\n\t\t\t\t},\n\t\t\t\tresolve: async ({ ctx, credential }): Promise<AuthResult | undefined> => {\n\t\t\t\t\tconst serverUrl = await resolveServerUrl(ctx, credential);\n\t\t\t\t\tif (!serverUrl) return undefined;\n\t\t\t\t\tconst apiKey = credential?.key ?? (await ctx.env(\"LLAMA_API_KEY\")) ?? \"local\";\n\t\t\t\t\treturn {\n\t\t\t\t\t\tauth: { apiKey, baseUrl: llamaInferenceUrl(serverUrl) },\n\t\t\t\t\t\tenv: { ...credential?.env, LLAMA_BASE_URL: serverUrl },\n\t\t\t\t\t\tsource: credential ? \"stored credential\" : \"LLAMA_BASE_URL\",\n\t\t\t\t\t};\n\t\t\t\t},\n\t\t\t},\n\t\t},\n\t\tgetModels: () => models,\n\t\tgetAllModels: () => [...models, ...classifiers],\n\t\trefreshModels: async (context: RefreshModelsContext): Promise<void> => {\n\t\t\tconst cachedContextWindows = new Map<string, number>();\n\t\t\tif (context.stored) {\n\t\t\t\tconst stored = context.stored.models.filter((model) => model.provider === LLAMA_PROVIDER_ID);\n\t\t\t\tconst restored = stored.filter(\n\t\t\t\t\t(model): model is Model<\"openai-completions\"> =>\n\t\t\t\t\t\tisModelType(model, \"chat\") && model.api === \"openai-completions\",\n\t\t\t\t);\n\t\t\t\tconst restoredClassifiers = stored.filter(\n\t\t\t\t\t(model): model is ClassifierModel<\"llama-cpp-classify\"> =>\n\t\t\t\t\t\tisModelType(model, \"classifier\") && model.api === \"llama-cpp-classify\",\n\t\t\t\t);\n\t\t\t\tfor (const model of [...restored, ...restoredClassifiers]) {\n\t\t\t\t\tcachedContextWindows.set(model.id, model.contextWindow);\n\t\t\t\t}\n\t\t\t\tif (\n\t\t\t\t\t!(await context.publish({\n\t\t\t\t\t\tupdate: () => {\n\t\t\t\t\t\t\tmodels = restored;\n\t\t\t\t\t\t\tclassifiers = restoredClassifiers;\n\t\t\t\t\t\t},\n\t\t\t\t\t}))\n\t\t\t\t) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (!context.allowNetwork || context.signal.aborted || context.credential?.type !== \"api_key\") return;\n\t\t\tconst serverUrl = credentialServerUrl(context.credential);\n\t\t\tif (!serverUrl) return;\n\t\t\tconst client = new LlamaClient(serverUrl, context.credential.key);\n\t\t\tconst catalog = await client.list({ signal: context.signal });\n\t\t\tif (context.signal.aborted) return;\n\t\t\tconst routerAutoload = await routerAutoloadEnabled(client, catalog, context.signal);\n\t\t\tif (context.signal.aborted) return;\n\t\t\tconst selectable = catalog.filter((model) => modelIsSelectable(model, routerAutoload));\n\t\t\tconst refreshed = await Promise.all(\n\t\t\t\tselectable.map(async (model) => {\n\t\t\t\t\tconst cachedContextWindow = cachedContextWindows.get(model.id);\n\t\t\t\t\t// Only loaded models expose their template without side effects. Unloaded autoload presets\n\t\t\t\t\t// would need to be loaded, while querying sleeping models may wake them. Those models remain\n\t\t\t\t\t// unclassified until they are loaded or woken and a later catalog refresh discovers them.\n\t\t\t\t\tif (model.status.value !== \"loaded\")\n\t\t\t\t\t\treturn toRelayModel(model, serverUrl, undefined, cachedContextWindow);\n\t\t\t\t\tconst props = await client.props({ model: model.id, signal: context.signal });\n\t\t\t\t\treturn toRelayModel(model, serverUrl, props, cachedContextWindow);\n\t\t\t\t}),\n\t\t\t);\n\t\t\tconst refreshedClassifiers = selectable.map((model) =>\n\t\t\t\ttoRelayClassifierModel(model, serverUrl, cachedContextWindows.get(model.id)),\n\t\t\t);\n\t\t\tif (context.signal.aborted) return;\n\t\t\tawait context.publish({\n\t\t\t\tpersist: { models: [...refreshed, ...refreshedClassifiers], checkedAt: Date.now() },\n\t\t\t\tupdate: () => {\n\t\t\t\t\tmodels = refreshed;\n\t\t\t\t\tclassifiers = refreshedClassifiers;\n\t\t\t\t},\n\t\t\t});\n\t\t},\n\t\tstream: (model, context, options) => stream(model, context, options as ProviderStreamOptions | undefined),\n\t\tstreamSimple: (model, context, options) => streamSimple(model, context, options),\n\t\tclassify: (model, context, options) => classifier.classify(model, context, options),\n\t};\n\n\treturn { provider, setCatalog };\n}\n"]}

package/dist/extensions/llama/ui.d.ts

@@ -0,0 +1,42 @@
+import type { ExtensionCommandContext } from "../../core/extensions/types.ts";
+import type { LlamaModelInfo, LlamaProgress } from "./client.ts";
+import type { HuggingFaceModel } from "./huggingface.ts";
+export type LlamaManagerAction = {
+    type: "model";
+    model: LlamaModelInfo;
+} | {
+    type: "download";
+} | {
+    type: "close";
+};
+interface ProgressState extends LlamaProgress {
+    title: string;
+    model: string;
+}
+export interface LlamaUi {
+    showModels(serverUrl: string, models: LlamaModelInfo[]): Promise<LlamaManagerAction>;
+    select(title: string, options: string[]): Promise<string | undefined>;
+    confirm(title: string, message: string): Promise<boolean>;
+    connectionError(serverUrl: string, message: string): Promise<"retry" | "close">;
+    searchModels(search: (query: string, signal: AbortSignal) => Promise<HuggingFaceModel[]>): Promise<string | undefined>;
+    showStatus(title: string, message: string): void;
+    progress(state: ProgressState): Promise<void>;
+    updateProgress(state: ProgressState): void;
+}
+export declare function showLlamaUi(ctx: ExtensionCommandContext, run: (ui: LlamaUi) => Promise<void>): Promise<void>;
+export declare function runWithProgress<T>(ui: LlamaUi, options: {
+    title: string;
+    model: string;
+    initialMessage: string;
+    cancelTitle: string;
+    cancelMessage: string;
+    run(signal: AbortSignal, update: (progress: LlamaProgress) => void): Promise<T>;
+    cancel(): Promise<void>;
+}): Promise<{
+    cancelled: true;
+} | {
+    cancelled: false;
+    value: T;
+}>;
+export {};
+//# sourceMappingURL=ui.d.ts.map

package/dist/extensions/llama/ui.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ui.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/ui.ts"],"names":[],"mappings":"AAcA,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,gCAAgC,CAAC;AAK9E,OAAO,KAAK,EAAE,cAAc,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AACjE,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AAIzD,MAAM,MAAM,kBAAkB,GAAG;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,KAAK,EAAE,cAAc,CAAA;CAAE,GAAG;IAAE,IAAI,EAAE,UAAU,CAAA;CAAE,GAAG;IAAE,IAAI,EAAE,OAAO,CAAA;CAAE,CAAC;AAErH,UAAU,aAAc,SAAQ,aAAa;IAC5C,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;CACd;AA+CD,MAAM,WAAW,OAAO;IACvB,UAAU,CAAC,SAAS,EAAE,MAAM,EAAE,MAAM,EAAE,cAAc,EAAE,GAAG,OAAO,CAAC,kBAAkB,CAAC,CAAC;IACrF,MAAM,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC,CAAC;IACtE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IAC1D,eAAe,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,GAAG,OAAO,CAAC,CAAC;IAChF,YAAY,CACX,MAAM,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,WAAW,KAAK,OAAO,CAAC,gBAAgB,EAAE,CAAC,GACzE,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC,CAAC;IAC/B,UAAU,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACjD,QAAQ,CAAC,KAAK,EAAE,aAAa,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC9C,cAAc,CAAC,KAAK,EAAE,aAAa,GAAG,IAAI,CAAC;CAC3C;AAwYD,wBAAsB,WAAW,CAAC,GAAG,EAAE,uBAAuB,EAAE,GAAG,EAAE,CAAC,EAAE,EAAE,OAAO,KAAK,OAAO,CAAC,IAAI,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,CAYlH;AAED,wBAAsB,eAAe,CAAC,CAAC,EACtC,EAAE,EAAE,OAAO,EACX,OAAO,EAAE;IACR,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,EAAE,MAAM,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,CAAC;IACtB,GAAG,CAAC,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,CAAC,QAAQ,EAAE,aAAa,KAAK,IAAI,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;IAChF,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;CACxB,GACC,OAAO,CAAC;IAAE,SAAS,EAAE,IAAI,CAAA;CAAE,GAAG;IAAE,SAAS,EAAE,KAAK,CAAC;IAAC,KAAK,EAAE,CAAC,CAAA;CAAE,CAAC,CAqC/D"}