Add MiniMax provider support (#656 by @dannote)

- Add minimax to KnownProvider and Api types
- Add MINIMAX_API_KEY to getEnvApiKey()
- Generate MiniMax-M2 and MiniMax-M2.1 models
- Add context overflow detection pattern
- Add tests to all required test files
- Update README and CHANGELOG with attribution

Also fixes:
- Bedrock duplicate toolResult ID when content has multiple blocks
- Sandbox extension unused parameter lint warning

packages/ai/CHANGELOG.md

@@ -4,6 +4,7 @@
 
 ### Added
 
+- MiniMax provider support with M2 and M2.1 models via Anthropic-compatible API ([#656](https://github.com/badlogic/pi-mono/pull/656) by [@dannote](https://github.com/dannote))
 - Add Amazon Bedrock provider with prompt caching for Claude models (experimental, tested with Anthropic Claude models only) ([#494](https://github.com/badlogic/pi-mono/pull/494) by [@unexge](https://github.com/unexge))
 - Added `serviceTier` option for OpenAI Responses requests ([#672](https://github.com/badlogic/pi-mono/pull/672) by [@markusylisiurunen](https://github.com/markusylisiurunen))
 - **Anthropic caching on OpenRouter**: Interactions with Anthropic models via OpenRouter now set a 5-minute cache point using Anthropic-style `cache_control` breakpoints on the last assistant or user message. ([#584](https://github.com/badlogic/pi-mono/pull/584) by [@nathyong](https://github.com/nathyong))

packages/ai/README.md

@@ -56,6 +56,7 @@ Unified LLM API with automatic model discovery, provider configuration, token an
 - **Cerebras**
 - **xAI**
 - **OpenRouter**
+- **MiniMax**
 - **GitHub Copilot** (requires OAuth, see below)
 - **Google Gemini CLI** (requires OAuth, see below)
 - **Antigravity** (requires OAuth, see below)
@@ -862,6 +863,7 @@ In Node.js environments, you can set environment variables to avoid passing API
 | xAI | `XAI_API_KEY` |
 | OpenRouter | `OPENROUTER_API_KEY` |
 | zAI | `ZAI_API_KEY` |
+| MiniMax | `MINIMAX_API_KEY` |
 | GitHub Copilot | `COPILOT_GITHUB_TOKEN` or `GH_TOKEN` or `GITHUB_TOKEN` |
 
 When set, the library automatically uses these keys:

packages/ai/scripts/generate-models.ts

@@ -478,6 +478,33 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 			}
 		}
 
+		// Process MiniMax models
+		if (data.minimax?.models) {
+			for (const [modelId, model] of Object.entries(data.minimax.models)) {
+				const m = model as ModelsDevModel;
+				if (m.tool_call !== true) continue;
+
+				models.push({
+					id: modelId,
+					name: m.name || modelId,
+					api: "anthropic-messages",
+					provider: "minimax",
+					// MiniMax's Anthropic-compatible API - SDK appends /v1/messages
+					baseUrl: "https://api.minimax.io/anthropic",
+					reasoning: m.reasoning === true,
+					input: m.modalities?.input?.includes("image") ? ["text", "image"] : ["text"],
+					cost: {
+						input: m.cost?.input || 0,
+						output: m.cost?.output || 0,
+						cacheRead: m.cost?.cache_read || 0,
+						cacheWrite: m.cost?.cache_write || 0,
+					},
+					contextWindow: m.limit?.context || 4096,
+					maxTokens: m.limit?.output || 4096,
+				});
+			}
+		}
+
 		console.log(`Loaded ${models.length} tool-capable models from models.dev`);
 		return models;
 	} catch (error) {

packages/ai/src/models.generated.ts

@@ -2686,6 +2686,42 @@ export const MODELS = {
 			maxTokens: 16384,
 		} satisfies Model<"openai-completions">,
 	},
+	"minimax": {
+		"MiniMax-M2": {
+			id: "MiniMax-M2",
+			name: "MiniMax-M2",
+			api: "anthropic-messages",
+			provider: "minimax",
+			baseUrl: "https://api.minimax.io/anthropic",
+			reasoning: true,
+			input: ["text"],
+			cost: {
+				input: 0.3,
+				output: 1.2,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 196608,
+			maxTokens: 128000,
+		} satisfies Model<"anthropic-messages">,
+		"MiniMax-M2.1": {
+			id: "MiniMax-M2.1",
+			name: "MiniMax-M2.1",
+			api: "anthropic-messages",
+			provider: "minimax",
+			baseUrl: "https://api.minimax.io/anthropic",
+			reasoning: true,
+			input: ["text"],
+			cost: {
+				input: 0.3,
+				output: 1.2,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+			contextWindow: 204800,
+			maxTokens: 131072,
+		} satisfies Model<"anthropic-messages">,
+	},
 	"mistral": {
 		"codestral-latest": {
 			id: "codestral-latest",
@@ -4529,7 +4565,7 @@ export const MODELS = {
 				cacheWrite: 18.75,
 			},
 			contextWindow: 200000,
-			maxTokens: 4096,
+			maxTokens: 32000,
 		} satisfies Model<"openai-completions">,
 		"anthropic/claude-opus-4.5": {
 			id: "anthropic/claude-opus-4.5",

packages/ai/src/providers/amazon-bedrock.ts

@@ -378,38 +378,34 @@ function convertMessages(context: Context, model: Model<"bedrock-converse-stream
 				// Bedrock requires all tool results to be in one message
 				const toolResults: ContentBlock.ToolResultMember[] = [];
 
-				// Add current tool result
-				for (const c of m.content) {
-					toolResults.push({
-						toolResult: {
-							toolUseId: m.toolCallId,
-							content: [
-								c.type === "image"
-									? { image: createImageBlock(c.mimeType, c.data) }
-									: { text: sanitizeSurrogates(c.text) },
-							],
-							status: m.isError ? ToolResultStatus.ERROR : ToolResultStatus.SUCCESS,
-						},
-					});
-				}
+				// Add current tool result with all content blocks combined
+				toolResults.push({
+					toolResult: {
+						toolUseId: m.toolCallId,
+						content: m.content.map((c) =>
+							c.type === "image"
+								? { image: createImageBlock(c.mimeType, c.data) }
+								: { text: sanitizeSurrogates(c.text) },
+						),
+						status: m.isError ? ToolResultStatus.ERROR : ToolResultStatus.SUCCESS,
+					},
+				});
 
 				// Look ahead for consecutive toolResult messages
 				let j = i + 1;
 				while (j < messages.length && messages[j].role === "toolResult") {
 					const nextMsg = messages[j] as ToolResultMessage;
-					for (const c of nextMsg.content) {
-						toolResults.push({
-							toolResult: {
-								toolUseId: nextMsg.toolCallId,
-								content: [
-									c.type === "image"
-										? { image: createImageBlock(c.mimeType, c.data) }
-										: { text: sanitizeSurrogates(c.text) },
-								],
-								status: nextMsg.isError ? ToolResultStatus.ERROR : ToolResultStatus.SUCCESS,
-							},
-						});
-					}
+					toolResults.push({
+						toolResult: {
+							toolUseId: nextMsg.toolCallId,
+							content: nextMsg.content.map((c) =>
+								c.type === "image"
+									? { image: createImageBlock(c.mimeType, c.data) }
+									: { text: sanitizeSurrogates(c.text) },
+							),
+							status: nextMsg.isError ? ToolResultStatus.ERROR : ToolResultStatus.SUCCESS,
+						},
+					});
 					j++;
 				}
 

packages/ai/src/stream.ts

@@ -98,6 +98,7 @@ export function getEnvApiKey(provider: any): string | undefined {
 		openrouter: "OPENROUTER_API_KEY",
 		zai: "ZAI_API_KEY",
 		mistral: "MISTRAL_API_KEY",
+		minimax: "MINIMAX_API_KEY",
 		opencode: "OPENCODE_API_KEY",
 	};
 

packages/ai/src/types.ts

@@ -58,6 +58,7 @@ export type KnownProvider =
 	| "openrouter"
 	| "zai"
 	| "mistral"
+	| "minimax"
 	| "opencode";
 export type Provider = KnownProvider | string;
 

packages/ai/src/utils/overflow.ts

@@ -17,6 +17,7 @@ import type { AssistantMessage } from "../types.js";
  * - llama.cpp: "the request exceeds the available context size, try increasing it"
  * - LM Studio: "tokens to keep from the initial prompt is greater than the context length"
  * - GitHub Copilot: "prompt token count of X exceeds the limit of Y"
+ * - MiniMax: "invalid params, context window exceeds limit"
  * - Cerebras: Returns "400 status code (no body)" - handled separately below
  * - Mistral: Returns "400 status code (no body)" - handled separately below
  * - z.ai: Does NOT error, accepts overflow silently - handled via usage.input > contextWindow
@@ -33,6 +34,7 @@ const OVERFLOW_PATTERNS = [
 	/exceeds the limit of \d+/i, // GitHub Copilot
 	/exceeds the available context size/i, // llama.cpp server
 	/greater than the context length/i, // LM Studio
+	/context window exceeds limit/i, // MiniMax
 	/context[_ ]length[_ ]exceeded/i, // Generic fallback
 	/too many tokens/i, // Generic fallback
 	/token limit exceeded/i, // Generic fallback

packages/ai/test/abort.test.ts

@@ -160,6 +160,18 @@ describe("AI Providers Abort Tests", () => {
 		});
 	});
 
+	describe.skipIf(!process.env.MINIMAX_API_KEY)("MiniMax Provider Abort", () => {
+		const llm = getModel("minimax", "MiniMax-M2.1");
+
+		it("should abort mid-stream", { retry: 3 }, async () => {
+			await testAbortSignal(llm);
+		});
+
+		it("should handle immediate abort", { retry: 3 }, async () => {
+			await testImmediateAbort(llm);
+		});
+	});
+
 	// Google Gemini CLI / Antigravity share the same provider, so one test covers both
 	describe("Google Gemini CLI Provider Abort", () => {
 		it.skipIf(!geminiCliToken)("should abort mid-stream", { retry: 3 }, async () => {

packages/ai/test/context-overflow.test.ts

@@ -396,6 +396,22 @@ describe("Context overflow error handling", () => {
 		}, 120000);
 	});
 
+	// =============================================================================
+	// MiniMax
+	// Expected pattern: TBD - need to test actual error message
+	// =============================================================================
+
+	describe.skipIf(!process.env.MINIMAX_API_KEY)("MiniMax", () => {
+		it("MiniMax-M2.1 - should detect overflow via isContextOverflow", async () => {
+			const model = getModel("minimax", "MiniMax-M2.1");
+			const result = await testContextOverflow(model, process.env.MINIMAX_API_KEY!);
+			logResult(result);
+
+			expect(result.stopReason).toBe("error");
+			expect(isContextOverflow(result.response, model.contextWindow)).toBe(true);
+		}, 120000);
+	});
+
 	// =============================================================================
 	// OpenRouter - Multiple backend providers
 	// Expected pattern: "maximum context length is X tokens"

packages/ai/test/empty.test.ts

@@ -322,6 +322,26 @@ describe("AI Providers Empty Message Tests", () => {
 		});
 	});
 
+	describe.skipIf(!process.env.MINIMAX_API_KEY)("MiniMax Provider Empty Messages", () => {
+		const llm = getModel("minimax", "MiniMax-M2.1");
+
+		it("should handle empty content array", { retry: 3, timeout: 30000 }, async () => {
+			await testEmptyMessage(llm);
+		});
+
+		it("should handle empty string content", { retry: 3, timeout: 30000 }, async () => {
+			await testEmptyStringMessage(llm);
+		});
+
+		it("should handle whitespace-only content", { retry: 3, timeout: 30000 }, async () => {
+			await testWhitespaceOnlyMessage(llm);
+		});
+
+		it("should handle empty assistant message in conversation", { retry: 3, timeout: 30000 }, async () => {
+			await testEmptyAssistantMessage(llm);
+		});
+	});
+
 	describe.skipIf(!hasBedrockCredentials())("Amazon Bedrock Provider Empty Messages", () => {
 		const llm = getModel("amazon-bedrock", "global.anthropic.claude-sonnet-4-5-20250929-v1:0");
 

packages/ai/test/stream.test.ts

@@ -699,6 +699,30 @@ describe("Generate E2E Tests", () => {
 		});
 	});
 
+	describe.skipIf(!process.env.MINIMAX_API_KEY)("MiniMax Provider (MiniMax-M2.1 via Anthropic Messages)", () => {
+		const llm = getModel("minimax", "MiniMax-M2.1");
+
+		it("should complete basic text generation", { retry: 3 }, async () => {
+			await basicTextGeneration(llm);
+		});
+
+		it("should handle tool calling", { retry: 3 }, async () => {
+			await handleToolCall(llm);
+		});
+
+		it("should handle streaming", { retry: 3 }, async () => {
+			await handleStreaming(llm);
+		});
+
+		it("should handle thinking mode", { retry: 3 }, async () => {
+			await handleThinking(llm, { thinkingEnabled: true, thinkingBudgetTokens: 2048 });
+		});
+
+		it("should handle multi-turn with thinking and tools", { retry: 3 }, async () => {
+			await multiTurn(llm, { thinkingEnabled: true, thinkingBudgetTokens: 2048 });
+		});
+	});
+
 	// =========================================================================
 	// OAuth-based providers (credentials from ~/.pi/agent/oauth.json)
 	// Tokens are resolved at module level (see oauthTokens above)

packages/ai/test/tokens.test.ts

@@ -46,7 +46,8 @@ async function testTokensOnAbort<TApi extends Api>(llm: Model<TApi>, options: Op
 	expect(msg.stopReason).toBe("aborted");
 
 	// OpenAI providers, OpenAI Codex, Gemini CLI, zai, Amazon Bedrock, and the GPT-OSS model on Antigravity only send usage in the final chunk,
-	// so when aborted they have no token stats Anthropic and Google send usage information early in the stream
+	// so when aborted they have no token stats. Anthropic and Google send usage information early in the stream.
+	// MiniMax reports input tokens but not output tokens when aborted.
 	if (
 		llm.api === "openai-completions" ||
 		llm.api === "openai-responses" ||
@@ -58,6 +59,10 @@ async function testTokensOnAbort<TApi extends Api>(llm: Model<TApi>, options: Op
 	) {
 		expect(msg.usage.input).toBe(0);
 		expect(msg.usage.output).toBe(0);
+	} else if (llm.provider === "minimax") {
+		// MiniMax reports input tokens early but output tokens only in final chunk
+		expect(msg.usage.input).toBeGreaterThan(0);
+		expect(msg.usage.output).toBe(0);
 	} else {
 		expect(msg.usage.input).toBeGreaterThan(0);
 		expect(msg.usage.output).toBeGreaterThan(0);
@@ -146,6 +151,14 @@ describe("Token Statistics on Abort", () => {
 		});
 	});
 
+	describe.skipIf(!process.env.MINIMAX_API_KEY)("MiniMax Provider", () => {
+		const llm = getModel("minimax", "MiniMax-M2.1");
+
+		it("should include token stats when aborted mid-stream", { retry: 3, timeout: 30000 }, async () => {
+			await testTokensOnAbort(llm);
+		});
+	});
+
 	// =========================================================================
 	// OAuth-based providers (credentials from ~/.pi/agent/oauth.json)
 	// =========================================================================

packages/ai/test/tool-call-without-result.test.ts

@@ -171,6 +171,14 @@ describe("Tool Call Without Result Tests", () => {
 		});
 	});
 
+	describe.skipIf(!process.env.MINIMAX_API_KEY)("MiniMax Provider", () => {
+		const model = getModel("minimax", "MiniMax-M2.1");
+
+		it("should filter out tool calls without corresponding tool results", { retry: 3, timeout: 30000 }, async () => {
+			await testToolCallWithoutResult(model);
+		});
+	});
+
 	describe.skipIf(!hasBedrockCredentials())("Amazon Bedrock Provider", () => {
 		const model = getModel("amazon-bedrock", "global.anthropic.claude-sonnet-4-5-20250929-v1:0");
 

packages/ai/test/total-tokens.test.ts

@@ -325,6 +325,29 @@ describe("totalTokens field", () => {
 		);
 	});
 
+	// =========================================================================
+	// MiniMax
+	// =========================================================================
+
+	describe.skipIf(!process.env.MINIMAX_API_KEY)("MiniMax", () => {
+		it(
+			"MiniMax-M2.1 - should return totalTokens equal to sum of components",
+			{ retry: 3, timeout: 60000 },
+			async () => {
+				const llm = getModel("minimax", "MiniMax-M2.1");
+
+				console.log(`\nMiniMax / ${llm.id}:`);
+				const { first, second } = await testTotalTokensWithCache(llm, { apiKey: process.env.MINIMAX_API_KEY });
+
+				logUsage("First request", first);
+				logUsage("Second request", second);
+
+				assertTotalTokensEqualsComponents(first);
+				assertTotalTokensEqualsComponents(second);
+			},
+		);
+	});
+
 	// =========================================================================
 	// OpenRouter - Multiple backend providers
 	// =========================================================================

packages/ai/test/unicode-surrogate.test.ts

@@ -618,6 +618,22 @@ describe("AI Providers Unicode Surrogate Pair Tests", () => {
 		});
 	});
 
+	describe.skipIf(!process.env.MINIMAX_API_KEY)("MiniMax Provider Unicode Handling", () => {
+		const llm = getModel("minimax", "MiniMax-M2.1");
+
+		it("should handle emoji in tool results", { retry: 3, timeout: 30000 }, async () => {
+			await testEmojiInToolResults(llm);
+		});
+
+		it("should handle real-world LinkedIn comment data with emoji", { retry: 3, timeout: 30000 }, async () => {
+			await testRealWorldLinkedInData(llm);
+		});
+
+		it("should handle unpaired high surrogate (0xD83D) in tool results", { retry: 3, timeout: 30000 }, async () => {
+			await testUnpairedHighSurrogate(llm);
+		});
+	});
+
 	describe.skipIf(!hasBedrockCredentials())("Amazon Bedrock Provider Unicode Handling", () => {
 		const llm = getModel("amazon-bedrock", "global.anthropic.claude-sonnet-4-5-20250929-v1:0");
 

packages/coding-agent/README.md

@@ -167,6 +167,7 @@ Add API keys to `~/.pi/agent/auth.json`:
 | xAI | `xai` | `XAI_API_KEY` |
 | OpenRouter | `openrouter` | `OPENROUTER_API_KEY` |
 | ZAI | `zai` | `ZAI_API_KEY` |
+| MiniMax | `minimax` | `MINIMAX_API_KEY` |
 
 Auth file keys take priority over environment variables.
 
@@ -1142,7 +1143,7 @@ pi [options] [@files...] [messages...]
 
 | Option | Description |
 |--------|-------------|
-| `--provider <name>` | Provider: `anthropic`, `openai`, `openai-codex`, `google`, `mistral`, `xai`, `groq`, `cerebras`, `openrouter`, `zai`, `github-copilot`, `google-gemini-cli`, `google-antigravity`, or custom |
+| `--provider <name>` | Provider: `anthropic`, `openai`, `openai-codex`, `google`, `google-vertex`, `amazon-bedrock`, `mistral`, `xai`, `groq`, `cerebras`, `openrouter`, `zai`, `minimax`, `github-copilot`, `google-gemini-cli`, `google-antigravity`, or custom |
 | `--model <id>` | Model ID |
 | `--api-key <key>` | API key (overrides environment) |
 | `--system-prompt <text\|file>` | Custom system prompt (text or file path) |

packages/coding-agent/examples/extensions/sandbox/index.ts

@@ -211,7 +211,7 @@ export default function (pi: ExtensionAPI) {
 	pi.registerTool({
 		...localBash,
 		label: "bash (sandboxed)",
-		async execute(id, params, onUpdate, ctx, signal) {
+		async execute(id, params, onUpdate, _ctx, signal) {
 			if (!sandboxEnabled || !sandboxInitialized) {
 				return localBash.execute(id, params, signal, onUpdate);
 			}

packages/coding-agent/src/cli/args.ts

@@ -243,6 +243,8 @@ ${chalk.bold("Environment Variables:")}
   XAI_API_KEY             - xAI Grok API key
   OPENROUTER_API_KEY      - OpenRouter API key
   ZAI_API_KEY             - ZAI API key
+  MISTRAL_API_KEY         - Mistral API key
+  MINIMAX_API_KEY         - MiniMax API key
   AWS_PROFILE             - AWS profile for Amazon Bedrock
   AWS_ACCESS_KEY_ID       - AWS access key for Amazon Bedrock
   AWS_SECRET_ACCESS_KEY   - AWS secret key for Amazon Bedrock

packages/coding-agent/src/core/model-resolver.ts

@@ -26,6 +26,7 @@ export const defaultModelPerProvider: Record<KnownProvider, string> = {
 	cerebras: "zai-glm-4.6",
 	zai: "glm-4.6",
 	mistral: "devstral-medium-latest",
+	minimax: "MiniMax-M2.1",
 	opencode: "claude-opus-4-5",
 };