feat: enhance model cycling with thinking levels and --thinking flag

PR #47 enhancements:
- Add thinking level syntax to --models (e.g., --models sonnet:high,haiku:low)
- First model in scope used as initial model when starting new session
- Auto-apply thinking level when cycling models with Ctrl+P
- Save both model and thinking to session AND settings for persistence
- Simplify UX by removing autoThinkingDisabled flag
- Fix model matching to prioritize exact matches over partial
- Support provider/modelId format (e.g., openrouter/openai/gpt-5.1-codex)

Issue #45:
- Add --thinking CLI flag to set thinking level directly
- Takes highest priority over all other thinking level sources

Closes #45

packages/coding-agent/CHANGELOG.md

@@ -5,9 +5,12 @@
 ### Added
 
 - **`/clear` Command**: New slash command to reset the conversation context and start a fresh session. Aborts any in-flight agent work, clears all messages, and creates a new session file. ([#48](https://github.com/badlogic/pi-mono/pull/48))
+- **Model Cycling with Thinking Levels**: The `--models` flag now supports thinking level syntax (e.g., `--models sonnet:high,haiku:low`). When cycling models with `Ctrl+P`, the associated thinking level is automatically applied. The first model in the scope is used as the initial model when starting a new session. Both model and thinking level changes are now saved to session and settings for persistence. ([#47](https://github.com/badlogic/pi-mono/pull/47))
+- **`--thinking` Flag**: New CLI flag to set thinking level directly (e.g., `--thinking high`). Valid values: `off`, `minimal`, `low`, `medium`, `high`. Takes highest priority over all other thinking level sources. ([#45](https://github.com/badlogic/pi-mono/issues/45))
 
 ### Fixed
 
+- **Model Matching Priority**: The `--models` flag now prioritizes exact matches over partial matches. Supports `provider/modelId` format (e.g., `openrouter/openai/gpt-5.1-codex`) for precise selection. Exact ID matches are tried before partial matching, so `--models gpt-5.1-codex` correctly selects `gpt-5.1-codex` instead of `openai/gpt-5.1-codex-mini`.
 - **Markdown Link Rendering**: Fixed links with identical text and href (e.g., `https://github.com/badlogic/pi-mono/pull/48/files`) being rendered twice. Now correctly compares raw text instead of styled text (which contains ANSI codes) when determining if link text matches href.
 
 ## [0.8.5] - 2025-11-21

packages/coding-agent/README.md

@@ -670,12 +670,27 @@ Continue the most recent session
 Select a session to resume (opens interactive selector)
 
 **--models <patterns>**
-Comma-separated model patterns for quick cycling with `Ctrl+P`. Patterns match against model IDs and names (case-insensitive). When multiple versions exist, prefers aliases over dated versions (e.g., `claude-sonnet-4-5` over `claude-sonnet-4-5-20250929`). Without this flag, `Ctrl+P` cycles through all available models.
+Comma-separated model patterns for quick cycling with `Ctrl+P`. Matching priority:
+1. `provider/modelId` exact match (e.g., `openrouter/openai/gpt-5.1-codex`)
+2. Exact model ID match (e.g., `gpt-5.1-codex`)
+3. Partial match against model IDs and names (case-insensitive)
+
+When multiple partial matches exist, prefers aliases over dated versions (e.g., `claude-sonnet-4-5` over `claude-sonnet-4-5-20250929`). Without this flag, `Ctrl+P` cycles through all available models.
+
+Each pattern can optionally include a thinking level suffix: `pattern:level` where level is one of `off`, `minimal`, `low`, `medium`, or `high`. When cycling models, the associated thinking level is automatically applied. The first model in the list is used as the initial model when starting a new session.
 
 Examples:
-- `--models claude-sonnet,gpt-4o` - Scope to Claude Sonnet and GPT-4o
-- `--models sonnet,haiku` - Match any model containing "sonnet" or "haiku"
-- `--models gemini` - All Gemini models
+- `--models openrouter/openai/gpt-5.1-codex` - Exact provider/model match
+- `--models gpt-5.1-codex` - Exact ID match (not `openai/gpt-5.1-codex-mini`)
+- `--models sonnet:high,haiku:low` - Sonnet with high thinking, Haiku with low thinking
+- `--models sonnet,haiku` - Partial match for any model containing "sonnet" or "haiku"
+
+**--thinking <level>**
+Set thinking level for reasoning-capable models. Valid values: `off`, `minimal`, `low`, `medium`, `high`. Takes highest priority over all other thinking level sources (saved settings, `--models` pattern levels, session restore).
+
+Examples:
+- `--thinking high` - Start with high thinking level
+- `--thinking off` - Disable thinking even if saved setting was different
 
 **--help, -h**
 Show help message
@@ -707,6 +722,13 @@ pi --provider openai --model gpt-4o "Help me refactor this code"
 # Limit model cycling to specific models
 pi --models claude-sonnet,claude-haiku,gpt-4o
 # Now Ctrl+P cycles only through those models
+
+# Model cycling with thinking levels
+pi --models sonnet:high,haiku:low
+# Starts with sonnet at high thinking, Ctrl+P switches to haiku at low thinking
+
+# Start with specific thinking level
+pi --thinking high "Solve this complex algorithm problem"
 ```
 
 ## Tools

packages/coding-agent/src/main.ts

@@ -39,6 +39,7 @@ interface Args {
 	model?: string;
 	apiKey?: string;
 	systemPrompt?: string;
+	thinking?: ThinkingLevel;
 	continue?: boolean;
 	resume?: boolean;
 	help?: boolean;
@@ -82,6 +83,17 @@ function parseArgs(args: string[]): Args {
 			result.session = args[++i];
 		} else if (arg === "--models" && i + 1 < args.length) {
 			result.models = args[++i].split(",").map((s) => s.trim());
+		} else if (arg === "--thinking" && i + 1 < args.length) {
+			const level = args[++i];
+			if (level === "off" || level === "minimal" || level === "low" || level === "medium" || level === "high") {
+				result.thinking = level;
+			} else {
+				console.error(
+					chalk.yellow(
+						`Warning: Invalid thinking level "${level}". Valid values: off, minimal, low, medium, high`,
+					),
+				);
+			}
 		} else if (!arg.startsWith("-")) {
 			result.messages.push(arg);
 		}
@@ -107,6 +119,7 @@ ${chalk.bold("Options:")}
   --session <path>        Use specific session file
   --no-session            Don't save session (ephemeral)
   --models <patterns>     Comma-separated model patterns for quick cycling with Ctrl+P
+  --thinking <level>      Set thinking level: off, minimal, low, medium, high
   --help, -h              Show this help
 
 ${chalk.bold("Examples:")}
@@ -131,6 +144,9 @@ ${chalk.bold("Examples:")}
   # Cycle models with fixed thinking levels
   pi --models sonnet:high,haiku:low
 
+  # Start with a specific thinking level
+  pi --thinking high "Solve this complex problem"
+
 ${chalk.bold("Environment Variables:")}
   ANTHROPIC_API_KEY       - Anthropic Claude API key
   ANTHROPIC_OAUTH_TOKEN   - Anthropic OAuth token (alternative to API key)
@@ -370,7 +386,38 @@ async function resolveModelScope(
 			}
 		}
 
-		// Find all models matching this pattern (case-insensitive partial match)
+		// Check for provider/modelId format (provider is everything before the first /)
+		const slashIndex = modelPattern.indexOf("/");
+		if (slashIndex !== -1) {
+			const provider = modelPattern.substring(0, slashIndex);
+			const modelId = modelPattern.substring(slashIndex + 1);
+			const providerMatch = availableModels.find(
+				(m) => m.provider.toLowerCase() === provider.toLowerCase() && m.id.toLowerCase() === modelId.toLowerCase(),
+			);
+			if (providerMatch) {
+				if (
+					!scopedModels.find(
+						(sm) => sm.model.id === providerMatch.id && sm.model.provider === providerMatch.provider,
+					)
+				) {
+					scopedModels.push({ model: providerMatch, thinkingLevel });
+				}
+				continue;
+			}
+			// No exact provider/model match - fall through to other matching
+		}
+
+		// Check for exact ID match (case-insensitive)
+		const exactMatch = availableModels.find((m) => m.id.toLowerCase() === modelPattern.toLowerCase());
+		if (exactMatch) {
+			// Exact match found - use it directly
+			if (!scopedModels.find((sm) => sm.model.id === exactMatch.id && sm.model.provider === exactMatch.provider)) {
+				scopedModels.push({ model: exactMatch, thinkingLevel });
+			}
+			continue;
+		}
+
+		// No exact match - fall back to partial matching
 		const matches = availableModels.filter(
 			(m) =>
 				m.id.toLowerCase().includes(modelPattern.toLowerCase()) ||
@@ -637,6 +684,12 @@ export async function main(args: string[]) {
 				process.exit(1);
 			}
 			initialModel = model;
+
+			// Also load saved thinking level if we're using saved model
+			const savedThinking = settingsManager.getDefaultThinkingLevel();
+			if (savedThinking) {
+				initialThinking = savedThinking;
+			}
 		}
 	}
 
@@ -769,6 +822,11 @@ export async function main(args: string[]) {
 		}
 	}
 
+	// CLI --thinking flag takes highest priority
+	if (parsed.thinking) {
+		initialThinking = parsed.thinking;
+	}
+
 	// Create agent (initialModel can be null in interactive mode)
 	const agent = new Agent({
 		initialState: {

packages/coding-agent/src/settings-manager.ts

@@ -6,6 +6,7 @@ export interface Settings {
 	lastChangelogVersion?: string;
 	defaultProvider?: string;
 	defaultModel?: string;
+	defaultThinkingLevel?: "off" | "minimal" | "low" | "medium" | "high";
 	queueMode?: "all" | "one-at-a-time";
 	theme?: string;
 }
@@ -98,4 +99,13 @@ export class SettingsManager {
 		this.settings.theme = theme;
 		this.save();
 	}
+
+	getDefaultThinkingLevel(): "off" | "minimal" | "low" | "medium" | "high" | undefined {
+		return this.settings.defaultThinkingLevel;
+	}
+
+	setDefaultThinkingLevel(level: "off" | "minimal" | "low" | "medium" | "high"): void {
+		this.settings.defaultThinkingLevel = level;
+		this.save();
+	}
 }

packages/coding-agent/src/tui/tui-renderer.ts

@@ -91,9 +91,6 @@ export class TuiRenderer {
 	// Model scope for quick cycling
 	private scopedModels: Array<{ model: Model<any>; thinkingLevel: ThinkingLevel }> = [];
 
-	// Track if user manually changed thinking (disables auto-thinking from model cycling)
-	private autoThinkingDisabled = false;
-
 	// Tool output expansion state
 	private toolOutputExpanded = false;
 
@@ -789,11 +786,9 @@ export class TuiRenderer {
 		// Apply the new thinking level
 		this.agent.setThinkingLevel(nextLevel);
 
-		// Disable auto-thinking since user manually changed it
-		this.autoThinkingDisabled = true;
-
-		// Save thinking level change to session
+		// Save thinking level change to session and settings
 		this.sessionManager.saveThinkingLevelChange(nextLevel);
+		this.settingsManager.setDefaultThinkingLevel(nextLevel);
 
 		// Update border color
 		this.updateEditorBorderColor();
@@ -840,17 +835,16 @@ export class TuiRenderer {
 			// Switch model
 			this.agent.setModel(nextModel);
 
-			// Apply thinking level if not disabled and model supports it
-			if (!this.autoThinkingDisabled && nextModel.reasoning) {
-				this.agent.setThinkingLevel(nextThinking);
-				this.sessionManager.saveThinkingLevelChange(nextThinking);
-				this.updateEditorBorderColor();
-			} else if (!this.autoThinkingDisabled && !nextModel.reasoning && nextThinking !== "off") {
-				// Model doesn't support thinking but user requested it - silently ignore
-				this.agent.setThinkingLevel("off");
-				this.sessionManager.saveThinkingLevelChange("off");
-				this.updateEditorBorderColor();
-			}
+			// Save model change to session and settings
+			this.sessionManager.saveModelChange(nextModel.provider, nextModel.id);
+			this.settingsManager.setDefaultModelAndProvider(nextModel.provider, nextModel.id);
+
+			// Apply thinking level (silently use "off" if model doesn't support thinking)
+			const effectiveThinking = nextModel.reasoning ? nextThinking : "off";
+			this.agent.setThinkingLevel(effectiveThinking);
+			this.sessionManager.saveThinkingLevelChange(effectiveThinking);
+			this.settingsManager.setDefaultThinkingLevel(effectiveThinking);
+			this.updateEditorBorderColor();
 
 			// Show notification
 			this.chatContainer.addChild(new Spacer(1));
@@ -874,7 +868,7 @@ export class TuiRenderer {
 
 			if (availableModels.length === 1) {
 				this.chatContainer.addChild(new Spacer(1));
-				this.chatContainer.addChild(new Text(theme.fg("dim", "Only one model in scope"), 1, 0));
+				this.chatContainer.addChild(new Text(theme.fg("dim", "Only one model available"), 1, 0));
 				this.ui.requestRender();
 				return;
 			}
@@ -902,6 +896,10 @@ export class TuiRenderer {
 			// Switch model
 			this.agent.setModel(nextModel);
 
+			// Save model change to session and settings
+			this.sessionManager.saveModelChange(nextModel.provider, nextModel.id);
+			this.settingsManager.setDefaultModelAndProvider(nextModel.provider, nextModel.id);
+
 			// Show notification
 			this.chatContainer.addChild(new Spacer(1));
 			this.chatContainer.addChild(new Text(theme.fg("dim", `Switched to ${nextModel.name || nextModel.id}`), 1, 0));
@@ -949,11 +947,9 @@ export class TuiRenderer {
 				// Apply the selected thinking level
 				this.agent.setThinkingLevel(level);
 
-				// Disable auto-thinking since user manually changed it
-				this.autoThinkingDisabled = true;
-
-				// Save thinking level change to session
+				// Save thinking level change to session and settings
 				this.sessionManager.saveThinkingLevelChange(level);
+				this.settingsManager.setDefaultThinkingLevel(level);
 
 				// Update border color
 				this.updateEditorBorderColor();
@@ -1107,9 +1103,6 @@ export class TuiRenderer {
 				// Apply the selected model
 				this.agent.setModel(model);
 
-				// Clear scoped models since user manually selected a model
-				this.scopedModels = [];
-
 				// Save model change to session
 				this.sessionManager.saveModelChange(model.provider, model.id);