gsd-pi 2.57.0 → 2.58.0-dev.778d6ac

package/src/resources/extensions/ollama/ollama-tool.ts

@@ -0,0 +1,218 @@
+// GSD2 — LLM-callable Ollama management tool
+/**
+ * Registers an ollama_manage tool that the LLM can call to interact
+ * with the local Ollama instance — list models, pull new ones, check status.
+ */
+
+import type { ExtensionAPI } from "@gsd/pi-coding-agent";
+import { Text } from "@gsd/pi-tui";
+import { Type } from "@sinclair/typebox";
+import * as client from "./ollama-client.js";
+import { discoverModels, formatModelForDisplay } from "./ollama-discovery.js";
+import { formatModelSize } from "./model-capabilities.js";
+
+interface OllamaToolDetails {
+	action: string;
+	model?: string;
+	modelCount?: number;
+	durationMs: number;
+	error?: string;
+}
+
+export function registerOllamaTool(pi: ExtensionAPI): void {
+	pi.registerTool({
+		name: "ollama_manage",
+		label: "Ollama",
+		description:
+			"Manage local Ollama models. List available models, pull new ones, " +
+			"check Ollama status, or see running models and resource usage. " +
+			"Use this when you need a specific local model that isn't available yet.",
+		promptSnippet: "Manage local Ollama models (list, pull, status, ps)",
+		promptGuidelines: [
+			"Use 'list' to see what models are available locally before trying to use one.",
+			"Use 'pull' to download a model that isn't available yet.",
+			"Use 'status' to check if Ollama is running.",
+			"Use 'ps' to see which models are loaded in memory and VRAM usage.",
+			"Common models: llama3.1:8b, qwen2.5-coder:7b, deepseek-r1:8b, codestral:22b",
+		],
+		parameters: Type.Object({
+			action: Type.Union(
+				[
+					Type.Literal("list"),
+					Type.Literal("pull"),
+					Type.Literal("status"),
+					Type.Literal("ps"),
+				],
+				{ description: "Action to perform" },
+			),
+			model: Type.Optional(
+				Type.String({ description: "Model name (required for pull)" }),
+			),
+		}),
+
+		async execute(_toolCallId, params, signal, onUpdate, _ctx) {
+			const startTime = Date.now();
+			const { action, model } = params;
+
+			try {
+				switch (action) {
+					case "status": {
+						const running = await client.isRunning();
+						if (!running) {
+							return {
+								content: [{ type: "text", text: "Ollama is not running. It needs to be started with 'ollama serve'." }],
+								details: { action, durationMs: Date.now() - startTime } as OllamaToolDetails,
+							};
+						}
+						const version = await client.getVersion();
+						return {
+							content: [{ type: "text", text: `Ollama${version ? ` v${version}` : ""} is running at ${client.getOllamaHost()}` }],
+							details: { action, durationMs: Date.now() - startTime } as OllamaToolDetails,
+						};
+					}
+
+					case "list": {
+						const running = await client.isRunning();
+						if (!running) {
+							return {
+								content: [{ type: "text", text: "Ollama is not running." }],
+								isError: true,
+								details: { action, durationMs: Date.now() - startTime, error: "not_running" } as OllamaToolDetails,
+							};
+						}
+
+						const models = await discoverModels();
+						if (models.length === 0) {
+							return {
+								content: [{ type: "text", text: "No models available. Pull one with action='pull'." }],
+								details: { action, modelCount: 0, durationMs: Date.now() - startTime } as OllamaToolDetails,
+							};
+						}
+
+						const lines = models.map((m) => formatModelForDisplay(m));
+						return {
+							content: [{ type: "text", text: `Available models:\n${lines.join("\n")}` }],
+							details: { action, modelCount: models.length, durationMs: Date.now() - startTime } as OllamaToolDetails,
+						};
+					}
+
+					case "pull": {
+						if (!model) {
+							return {
+								content: [{ type: "text", text: "Error: 'model' parameter is required for pull action." }],
+								isError: true,
+								details: { action, durationMs: Date.now() - startTime, error: "missing_model" } as OllamaToolDetails,
+							};
+						}
+
+						const running = await client.isRunning();
+						if (!running) {
+							return {
+								content: [{ type: "text", text: "Ollama is not running." }],
+								isError: true,
+								details: { action, model, durationMs: Date.now() - startTime, error: "not_running" } as OllamaToolDetails,
+							};
+						}
+
+						let lastStatus = "";
+						await client.pullModel(model, (progress) => {
+							if (progress.total && progress.completed) {
+								const pct = Math.floor((progress.completed / progress.total) * 100);
+								const status = `Pulling ${model}... ${pct}%`;
+								if (status !== lastStatus) {
+									lastStatus = status;
+									onUpdate?.({ content: [{ type: "text", text: status }], details: { action, model, durationMs: Date.now() - startTime } as OllamaToolDetails });
+								}
+							} else if (progress.status && progress.status !== lastStatus) {
+								lastStatus = progress.status;
+								onUpdate?.({ content: [{ type: "text", text: `${model}: ${progress.status}` }], details: { action, model, durationMs: Date.now() - startTime } as OllamaToolDetails });
+							}
+						}, signal);
+
+						return {
+							content: [{ type: "text", text: `Successfully pulled ${model}` }],
+							details: { action, model, durationMs: Date.now() - startTime } as OllamaToolDetails,
+						};
+					}
+
+					case "ps": {
+						const running = await client.isRunning();
+						if (!running) {
+							return {
+								content: [{ type: "text", text: "Ollama is not running." }],
+								isError: true,
+								details: { action, durationMs: Date.now() - startTime, error: "not_running" } as OllamaToolDetails,
+							};
+						}
+
+						const ps = await client.getRunningModels();
+						if (!ps.models || ps.models.length === 0) {
+							return {
+								content: [{ type: "text", text: "No models currently loaded in memory." }],
+								details: { action, modelCount: 0, durationMs: Date.now() - startTime } as OllamaToolDetails,
+							};
+						}
+
+						const lines = ps.models.map((m) => {
+							const vram = m.size_vram > 0 ? `${formatModelSize(m.size_vram)} VRAM` : "CPU";
+							return `${m.name} — ${formatModelSize(m.size)} total, ${vram}`;
+						});
+
+						return {
+							content: [{ type: "text", text: `Loaded models:\n${lines.join("\n")}` }],
+							details: { action, modelCount: ps.models.length, durationMs: Date.now() - startTime } as OllamaToolDetails,
+						};
+					}
+
+					default:
+						return {
+							content: [{ type: "text", text: `Unknown action: ${action}` }],
+							isError: true,
+							details: { action, durationMs: Date.now() - startTime, error: "unknown_action" } as OllamaToolDetails,
+						};
+				}
+			} catch (err) {
+				const msg = err instanceof Error ? err.message : String(err);
+				return {
+					content: [{ type: "text", text: `Ollama error: ${msg}` }],
+					isError: true,
+					details: { action, model, durationMs: Date.now() - startTime, error: msg } as OllamaToolDetails,
+				};
+			}
+		},
+
+		renderCall(args, theme) {
+			let text = theme.fg("toolTitle", theme.bold("ollama "));
+			text += theme.fg("accent", args.action);
+			if (args.model) {
+				text += theme.fg("dim", ` ${args.model}`);
+			}
+			return new Text(text, 0, 0);
+		},
+
+		renderResult(result, { isPartial, expanded }, theme) {
+			const d = result.details as OllamaToolDetails | undefined;
+
+			if (isPartial) return new Text(theme.fg("warning", "Working..."), 0, 0);
+			if ((result as any).isError || d?.error) {
+				return new Text(theme.fg("error", `Error: ${d?.error ?? "unknown"}`), 0, 0);
+			}
+
+			let text = theme.fg("success", d?.action ?? "done");
+			if (d?.modelCount !== undefined) {
+				text += theme.fg("dim", ` (${d.modelCount} models)`);
+			}
+			text += theme.fg("dim", ` ${d?.durationMs ?? 0}ms`);
+
+			if (expanded) {
+				const content = result.content[0];
+				if (content?.type === "text") {
+					const preview = content.text.split("\n").slice(0, 10).join("\n");
+					text += "\n\n" + theme.fg("dim", preview);
+				}
+			}
+
+			return new Text(text, 0, 0);
+		},
+	});
+}

package/src/resources/extensions/ollama/tests/model-capabilities.test.ts

@@ -0,0 +1,162 @@
+// GSD2 — Tests for Ollama model capability detection
+import { describe, it } from "node:test";
+import assert from "node:assert/strict";
+import {
+	getModelCapabilities,
+	estimateContextFromParams,
+	humanizeModelName,
+	formatModelSize,
+} from "../model-capabilities.js";
+
+// ─── getModelCapabilities ────────────────────────────────────────────────────
+
+describe("getModelCapabilities", () => {
+	it("returns reasoning for deepseek-r1 models", () => {
+		const caps = getModelCapabilities("deepseek-r1:8b");
+		assert.equal(caps.reasoning, true);
+		assert.equal(caps.contextWindow, 131072);
+	});
+
+	it("returns reasoning for qwq models", () => {
+		const caps = getModelCapabilities("qwq:32b");
+		assert.equal(caps.reasoning, true);
+	});
+
+	it("returns vision for llava models", () => {
+		const caps = getModelCapabilities("llava:7b");
+		assert.deepEqual(caps.input, ["text", "image"]);
+	});
+
+	it("returns vision for llama3.2-vision models", () => {
+		const caps = getModelCapabilities("llama3.2-vision:11b");
+		assert.deepEqual(caps.input, ["text", "image"]);
+	});
+
+	it("returns correct context for llama3.1", () => {
+		const caps = getModelCapabilities("llama3.1:8b");
+		assert.equal(caps.contextWindow, 131072);
+	});
+
+	it("returns correct context for llama3 (no .1)", () => {
+		const caps = getModelCapabilities("llama3:8b");
+		assert.equal(caps.contextWindow, 8192);
+	});
+
+	it("returns correct context for llama2", () => {
+		const caps = getModelCapabilities("llama2:7b");
+		assert.equal(caps.contextWindow, 4096);
+	});
+
+	it("returns correct context for qwen2.5-coder", () => {
+		const caps = getModelCapabilities("qwen2.5-coder:7b");
+		assert.equal(caps.contextWindow, 131072);
+		assert.equal(caps.maxTokens, 32768);
+	});
+
+	it("returns correct context for codestral", () => {
+		const caps = getModelCapabilities("codestral:22b");
+		assert.equal(caps.contextWindow, 262144);
+	});
+
+	it("returns correct context for mistral-nemo", () => {
+		const caps = getModelCapabilities("mistral-nemo:12b");
+		assert.equal(caps.contextWindow, 131072);
+	});
+
+	it("returns correct context for gemma3", () => {
+		const caps = getModelCapabilities("gemma3:9b");
+		assert.equal(caps.contextWindow, 131072);
+	});
+
+	it("returns empty object for unknown models", () => {
+		const caps = getModelCapabilities("totally-unknown-model:3b");
+		assert.deepEqual(caps, {});
+	});
+
+	it("strips tag before matching", () => {
+		const caps = getModelCapabilities("llama3.1:70b-instruct-q4_0");
+		assert.equal(caps.contextWindow, 131072);
+	});
+
+	it("matches case-insensitively", () => {
+		const caps = getModelCapabilities("Llama3.1:8B");
+		assert.equal(caps.contextWindow, 131072);
+	});
+});
+
+// ─── estimateContextFromParams ───────────────────────────────────────────────
+
+describe("estimateContextFromParams", () => {
+	it("estimates 8192 for small models", () => {
+		assert.equal(estimateContextFromParams("1.5B"), 8192);
+	});
+
+	it("estimates 16384 for 7B models", () => {
+		assert.equal(estimateContextFromParams("7B"), 16384);
+	});
+
+	it("estimates 32768 for 13B models", () => {
+		assert.equal(estimateContextFromParams("13B"), 32768);
+	});
+
+	it("estimates 65536 for 34B models", () => {
+		assert.equal(estimateContextFromParams("34B"), 65536);
+	});
+
+	it("estimates 131072 for 70B+ models", () => {
+		assert.equal(estimateContextFromParams("70B"), 131072);
+	});
+
+	it("handles decimal sizes", () => {
+		assert.equal(estimateContextFromParams("7.5B"), 16384);
+	});
+
+	it("handles M (millions)", () => {
+		assert.equal(estimateContextFromParams("500M"), 8192);
+	});
+
+	it("returns 8192 for unparseable input", () => {
+		assert.equal(estimateContextFromParams("unknown"), 8192);
+	});
+
+	it("returns 8192 for empty string", () => {
+		assert.equal(estimateContextFromParams(""), 8192);
+	});
+});
+
+// ─── humanizeModelName ───────────────────────────────────────────────────────
+
+describe("humanizeModelName", () => {
+	it("capitalizes and adds tag", () => {
+		assert.equal(humanizeModelName("llama3.1:8b"), "Llama 3.1 8B");
+	});
+
+	it("handles latest tag", () => {
+		assert.equal(humanizeModelName("llama3.1:latest"), "Llama 3.1");
+	});
+
+	it("handles no tag", () => {
+		assert.equal(humanizeModelName("llama3.1"), "Llama 3.1");
+	});
+
+	it("handles hyphenated names", () => {
+		const result = humanizeModelName("deepseek-r1:8b");
+		assert.ok(result.includes("8B"));
+	});
+});
+
+// ─── formatModelSize ─────────────────────────────────────────────────────────
+
+describe("formatModelSize", () => {
+	it("formats GB", () => {
+		assert.equal(formatModelSize(4_700_000_000), "4.7 GB");
+	});
+
+	it("formats MB", () => {
+		assert.equal(formatModelSize(500_000_000), "500.0 MB");
+	});
+
+	it("formats KB", () => {
+		assert.equal(formatModelSize(500_000), "500 KB");
+	});
+});

package/src/resources/extensions/ollama/tests/ollama-client.test.ts

@@ -0,0 +1,38 @@
+// GSD2 — Tests for Ollama HTTP client
+import { describe, it, beforeEach, afterEach } from "node:test";
+import assert from "node:assert/strict";
+import { getOllamaHost } from "../ollama-client.js";
+
+// ─── getOllamaHost ──────────────────────────────────────────────────────────
+
+describe("getOllamaHost", () => {
+	const originalHost = process.env.OLLAMA_HOST;
+
+	afterEach(() => {
+		if (originalHost === undefined) {
+			delete process.env.OLLAMA_HOST;
+		} else {
+			process.env.OLLAMA_HOST = originalHost;
+		}
+	});
+
+	it("returns default when OLLAMA_HOST is not set", () => {
+		delete process.env.OLLAMA_HOST;
+		assert.equal(getOllamaHost(), "http://localhost:11434");
+	});
+
+	it("returns OLLAMA_HOST when set with scheme", () => {
+		process.env.OLLAMA_HOST = "http://myhost:12345";
+		assert.equal(getOllamaHost(), "http://myhost:12345");
+	});
+
+	it("adds http:// when OLLAMA_HOST has no scheme", () => {
+		process.env.OLLAMA_HOST = "myhost:12345";
+		assert.equal(getOllamaHost(), "http://myhost:12345");
+	});
+
+	it("preserves https:// scheme", () => {
+		process.env.OLLAMA_HOST = "https://secure-ollama.example.com";
+		assert.equal(getOllamaHost(), "https://secure-ollama.example.com");
+	});
+});

package/src/resources/extensions/ollama/tests/ollama-discovery.test.ts

@@ -0,0 +1,28 @@
+// GSD2 — Tests for Ollama model discovery and enrichment
+import { describe, it, afterEach } from "node:test";
+import assert from "node:assert/strict";
+import { getOllamaOpenAIBaseUrl } from "../ollama-discovery.js";
+
+// ─── getOllamaOpenAIBaseUrl ─────────────────────────────────────────────────
+
+describe("getOllamaOpenAIBaseUrl", () => {
+	const originalHost = process.env.OLLAMA_HOST;
+
+	afterEach(() => {
+		if (originalHost === undefined) {
+			delete process.env.OLLAMA_HOST;
+		} else {
+			process.env.OLLAMA_HOST = originalHost;
+		}
+	});
+
+	it("returns default OpenAI-compat URL", () => {
+		delete process.env.OLLAMA_HOST;
+		assert.equal(getOllamaOpenAIBaseUrl(), "http://localhost:11434/v1");
+	});
+
+	it("appends /v1 to custom OLLAMA_HOST", () => {
+		process.env.OLLAMA_HOST = "http://remote:9999";
+		assert.equal(getOllamaOpenAIBaseUrl(), "http://remote:9999/v1");
+	});
+});

package/src/resources/extensions/ollama/types.ts

@@ -0,0 +1,130 @@
+// GSD2 — Ollama API response types
+
+/**
+ * Type definitions for the Ollama REST API.
+ * Reference: https://github.com/ollama/ollama/blob/main/docs/api.md
+ */
+
+// ─── /api/tags ──────────────────────────────────────────────────────────────
+
+export interface OllamaModelDetails {
+	parent_model: string;
+	format: string;
+	family: string;
+	families: string[] | null;
+	parameter_size: string;
+	quantization_level: string;
+}
+
+export interface OllamaModelInfo {
+	name: string;
+	model: string;
+	modified_at: string;
+	size: number;
+	digest: string;
+	details: OllamaModelDetails;
+}
+
+export interface OllamaTagsResponse {
+	models: OllamaModelInfo[];
+}
+
+// ─── /api/show ──────────────────────────────────────────────────────────────
+
+export interface OllamaShowResponse {
+	modelfile: string;
+	parameters: string;
+	template: string;
+	details: OllamaModelDetails;
+	model_info: Record<string, unknown>;
+}
+
+// ─── /api/ps ────────────────────────────────────────────────────────────────
+
+export interface OllamaRunningModel {
+	name: string;
+	model: string;
+	size: number;
+	digest: string;
+	details: OllamaModelDetails;
+	expires_at: string;
+	size_vram: number;
+}
+
+export interface OllamaPsResponse {
+	models: OllamaRunningModel[];
+}
+
+// ─── /api/pull ──────────────────────────────────────────────────────────────
+
+export interface OllamaPullProgress {
+	status: string;
+	digest?: string;
+	total?: number;
+	completed?: number;
+}
+
+// ─── /api/version ───────────────────────────────────────────────────────────
+
+export interface OllamaVersionResponse {
+	version: string;
+}
+
+// ─── /api/chat ──────────────────────────────────────────────────────────────
+
+export interface OllamaChatMessage {
+	role: "system" | "user" | "assistant" | "tool";
+	content: string;
+	images?: string[];
+	tool_calls?: OllamaToolCall[];
+}
+
+export interface OllamaToolCall {
+	function: {
+		name: string;
+		arguments: Record<string, unknown>;
+	};
+}
+
+export interface OllamaTool {
+	type: "function";
+	function: {
+		name: string;
+		description: string;
+		parameters: {
+			type: "object";
+			required?: string[];
+			properties: Record<string, unknown>;
+		};
+	};
+}
+
+export interface OllamaChatRequest {
+	model: string;
+	messages: OllamaChatMessage[];
+	stream?: boolean;
+	tools?: OllamaTool[];
+	options?: {
+		num_ctx?: number;
+		num_predict?: number;
+		temperature?: number;
+		top_p?: number;
+		top_k?: number;
+		stop?: string[];
+	};
+	keep_alive?: string;
+}
+
+export interface OllamaChatResponse {
+	model: string;
+	created_at: string;
+	message: OllamaChatMessage;
+	done: boolean;
+	done_reason?: string;
+	total_duration?: number;
+	load_duration?: number;
+	prompt_eval_count?: number;
+	prompt_eval_duration?: number;
+	eval_count?: number;
+	eval_duration?: number;
+}

package/src/resources/extensions/search-the-web/extension-manifest.json

@@ -8,6 +8,6 @@
   "provides": {
     "tools": ["search-the-web", "fetch_page", "search_and_read", "web_search"],
     "commands": ["search-provider"],
-    "hooks": ["model_select", "before_provider_request"]
+    "hooks": ["session_start", "model_select", "before_provider_request"]
   }
 }

package/src/resources/extensions/shared/interview-ui.ts

@@ -105,7 +105,7 @@ export interface WrapUpOptions {
 // ─── Constants ────────────────────────────────────────────────────────────────
 
 const OTHER_OPTION_LABEL = "None of the above";
-const OTHER_OPTION_DESCRIPTION = "Press TAB to add optional notes.";
+const OTHER_OPTION_DESCRIPTION = "Select to type your own answer.";
 
 // ─── Wrap-up screen ───────────────────────────────────────────────────────────
 
@@ -295,6 +295,17 @@ export async function showInterviewRound(
 				states[currentIdx].committedIndex = states[currentIdx].cursorIndex;
 			}
 
+			// Auto-open the notes field when "None of the above" is selected
+			// so the user can immediately provide a free-text explanation
+			// instead of being trapped in a re-asking loop (bug #2715).
+			if (!isMultiSelect(currentIdx) && states[currentIdx].cursorIndex === noneOrDoneIdx(currentIdx)) {
+				states[currentIdx].notesVisible = true;
+				focusNotes = true;
+				loadStateToEditor();
+				refresh();
+				return;
+			}
+
 			if (isMultiQuestion && currentIdx < questions.length - 1) {
 				let next = currentIdx + 1;
 				for (let i = 0; i < questions.length; i++) {