pi-incoai 0.1.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Sergiu Truta
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,99 @@
+# pi-incoai
+
+[Inco AI](https://inco.ai) (`inco.ai`) model provider extension for the
+[Pi coding agent](https://pi.dev).
+
+Inco serves open-weight models (Kimi K3, GLM 5.3, MiniMax M3, DeepSeek V4.1, …)
+over an OpenAI-compatible API at `https://api.inco.ai/v1`. This package registers
+the `incoai` provider, adds the public model catalog, and refreshes it from
+`GET /v1/models` so workspace/private models show up too.
+
+## Install
+
+```bash
+pi install npm:pi-incoai          # from npm
+pi install git:github.com/serggiu/pi-incoai   # from git
+pi install /path/to/pi-incoai     # from a local checkout
+```
+
+or try it for one run without installing:
+
+```bash
+pi -e /path/to/pi-incoai
+```
+
+## Authenticate
+
+1. Create an API key at <https://platform.inco.ai/keys> (it looks like `sk-inco-…`).
+2. Either set the environment variable in the shell that starts Pi:
+
+   ```bash
+   export INCO_API_KEY=sk-inco-...
+   ```
+
+   or sign in interactively:
+
+   ```
+   /login
+   ```
+
+   Pick **incoai** → *Sign in with an API key* → paste the key. Pi verifies the
+   key against `GET /v1/models` before saving it, then refreshes the model
+   catalog. Credentials are stored in `~/.pi/agent/auth.json`.
+
+## Use
+
+```
+/model            # search for "incoai"
+```
+
+```bash
+pi --model incoai/kimi-k3 "…"
+pi --model incoai/glm-5.3:fast "…"
+```
+
+Model IDs keep their `:fast` suffix; Pi resolves the full id before treating a
+trailing `:level` as a thinking level.
+
+## Models
+
+The bundled baseline catalog is replaced by the live catalog on refresh. Pricing,
+context length, and reasoning support come from `GET /v1/models`.
+
+| Model | Context | Max output | Image input |
+|---|---|---|---|
+| `kimi-k3` | 1.0M | 131K | yes |
+| `kimi-k3:fast` | 1.0M | 131K | yes |
+| `deepseek-v4.1-flash` | 1.0M | 384K | yes |
+| `deepseek-v4.1-flash:fast` | 1.0M | 384K | yes |
+| `glm-5.3` | 1.0M | 131K | no |
+| `glm-5.3:fast` | 1.0M | 131K | no |
+| `glm-5.3-flash` | 1.0M | 131K | yes |
+| `glm-5.3-flash:fast` | 1.0M | 131K | yes |
+| `minimax-m3` | 1.0M | 512K | yes |
+| `minimax-m3:fast` | 1.0M | 512K | yes |
+
+## Notes
+
+- Inco is prepaid: requests return `402` when the balance is empty.
+- Inco exposes both OpenAI Chat Completions and Anthropic Messages. This
+  extension uses Chat Completions, the surface its catalog metadata
+  (`reasoning_effort`, `reasoning_content`) is documented for.
+- Models that advertise `capabilities.reasoning_effort` receive `reasoning_effort`.
+  On the others Pi still surfaces streamed `reasoning_content` but sends no effort.
+- `--list-models` and `pi auth check` do not load extensions, so they show the
+  bundled baseline catalog only. Opening `/model` triggers the live refresh.
+
+## Development
+
+```text
+extensions/pi-incoai/index.ts   # the provider extension
+```
+
+Pi loads extensions from the `extensions/` directory of a package. Edit the file
+and run `/reload` in an active session, or restart Pi. To exercise the catalog
+mapping without a key, `GET https://api.inco.ai/v1/models` is public.
+
+## License
+
+MIT

package/extensions/pi-incoai/index.ts

@@ -0,0 +1,225 @@
+/**
+ * Inco AI (inco.ai) provider extension for the Pi coding agent.
+ *
+ * Inco serves open-weight models (Kimi K3, GLM 5.3, MiniMax M3, DeepSeek V4.1,
+ * …) through an OpenAI-compatible API at https://api.inco.ai/v1.
+ *
+ * This extension:
+ * - registers the `incoai` provider using pi-ai's OpenAI Chat Completions implementation;
+ * - authenticates with `INCO_API_KEY` or `pi`'s `/login incoai` flow, verifying the
+ *   key against `GET /v1/models` before it is stored;
+ * - ships a baseline catalog of the public models, refreshed from `GET /v1/models`
+ *   (the live catalog also includes private/workspace models when a key is configured).
+ */
+
+import {
+	createProvider,
+	envApiKeyAuth,
+	openAICompletionsApi,
+	type ApiKeyAuth,
+	type Model,
+	type RefreshModelsContext,
+} from "@earendil-works/pi-ai/compat";
+import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
+
+const PROVIDER_ID = "incoai";
+const PROVIDER_NAME = "incoai";
+const BASE_URL = "https://api.inco.ai/v1";
+const MODELS_URL = `${BASE_URL}/models`;
+
+const DEFAULT_CONTEXT_WINDOW = 1_000_000;
+const DEFAULT_MAX_TOKENS = 131_072;
+
+/** One entry of `GET /v1/models` (fields are optional; the API may omit them). */
+interface IncoCatalogEntry {
+	id: string;
+	name?: string;
+	context_length?: number;
+	pricing?: {
+		input?: number;
+		cached_input?: number;
+		output?: number;
+	};
+	capabilities?: {
+		reasoning_effort?: boolean;
+	};
+	modalities?: {
+		input?: string[];
+		output?: string[];
+	};
+}
+
+/**
+ * Metadata the Inco catalog does not report, kept per model id.
+ *
+ * `maxTokens` and text/image support are not part of `GET /v1/models`, so this
+ * supplies conservative values and fills gaps when `modalities` is absent.
+ */
+interface CuratedModel {
+	maxTokens: number;
+	input: ("text" | "image")[];
+}
+
+const CURATED: Record<string, CuratedModel> = {
+	"kimi-k3": { maxTokens: 131_072, input: ["text", "image"] },
+	"kimi-k3:fast": { maxTokens: 131_072, input: ["text", "image"] },
+	"deepseek-v4.1-flash": { maxTokens: 384_000, input: ["text", "image"] },
+	"deepseek-v4.1-flash:fast": { maxTokens: 384_000, input: ["text", "image"] },
+	"glm-5.3": { maxTokens: 131_072, input: ["text"] },
+	"glm-5.3:fast": { maxTokens: 131_072, input: ["text"] },
+	"glm-5.3-flash": { maxTokens: 131_072, input: ["text", "image"] },
+	"glm-5.3-flash:fast": { maxTokens: 131_072, input: ["text", "image"] },
+	"minimax-m3": { maxTokens: 512_000, input: ["text", "image"] },
+	"minimax-m3:fast": { maxTokens: 512_000, input: ["text", "image"] },
+};
+
+/**
+ * Snapshot of the public catalog, used as the baseline model list so models
+ * appear immediately. A successful `GET /v1/models` refresh replaces entries
+ * with the same id and adds any workspace-private models.
+ */
+const CATALOG_SNAPSHOT: IncoCatalogEntry[] = [
+	{ id: "kimi-k3", name: "Kimi K3", context_length: 1048576, pricing: { input: 3, cached_input: 0.3, output: 15 }, capabilities: { reasoning_effort: true } },
+	{ id: "deepseek-v4.1-flash", name: "DeepSeek V4.1 Flash", context_length: 1048576, pricing: { input: 0.3, cached_input: 0.006, output: 1.2 }, capabilities: { reasoning_effort: true } },
+	{ id: "glm-5.3-flash", name: "GLM 5.3 Flash", context_length: 1048576, pricing: { input: 0.15, cached_input: 0.03, output: 0.5 }, capabilities: { reasoning_effort: false } },
+	{ id: "kimi-k3:fast", name: "Kimi K3 (Fast)", context_length: 1048576, pricing: { input: 6, cached_input: 0.6, output: 30 }, capabilities: { reasoning_effort: false } },
+	{ id: "deepseek-v4.1-flash:fast", name: "DeepSeek V4.1 Flash (Fast)", context_length: 1048576, pricing: { input: 0.6, cached_input: 0.012, output: 2.4 }, capabilities: { reasoning_effort: false }, modalities: { input: ["text", "image"] } },
+	{ id: "glm-5.3", name: "GLM 5.3", context_length: 1048576, pricing: { input: 1.4, cached_input: 0.26, output: 4.4 }, capabilities: { reasoning_effort: true } },
+	{ id: "glm-5.3:fast", name: "GLM 5.3 (Fast)", context_length: 1048576, pricing: { input: 2.8, cached_input: 0.52, output: 8.8 }, capabilities: { reasoning_effort: false } },
+	{ id: "glm-5.3-flash:fast", name: "GLM 5.3 Flash (Fast)", context_length: 1048576, pricing: { input: 0.15, cached_input: 0.03, output: 0.5 }, capabilities: { reasoning_effort: false }, modalities: { input: ["text", "image"] } },
+	{ id: "minimax-m3", name: "MiniMax M3", context_length: 1048576, pricing: { input: 0.3, cached_input: 0.06, output: 1.2 }, capabilities: { reasoning_effort: true }, modalities: { input: ["text", "image"] } },
+	{ id: "minimax-m3:fast", name: "MiniMax M3 (Fast)", context_length: 1048576, pricing: { input: 0.6, cached_input: 0.12, output: 2.4 }, capabilities: { reasoning_effort: false }, modalities: { input: ["text", "image"] } },
+];
+
+/** Build an error from a failed catalog response, preferring Inco's error envelope. */
+async function catalogError(response: Response): Promise<Error> {
+	const status = `HTTP ${response.status} ${response.statusText}`.trim();
+	let message = `Inco model catalog request failed: ${status}`;
+	try {
+		const body = (await response.json()) as { error?: { message?: string; code?: string } };
+		const detail = body?.error?.message;
+		const code = body?.error?.code;
+		if (detail) {
+			message = `Inco model catalog request failed: ${status}${code ? ` (${code})` : ""} - ${detail}`;
+		}
+	} catch {
+		// Non-JSON error body; keep the status line.
+	}
+	return new Error(message);
+}
+
+/** Authenticated (or public) catalog request, used by both login and refresh. */
+function requestCatalog(apiKey: string | undefined, signal: AbortSignal): Promise<Response> {
+	const headers: Record<string, string> = { Accept: "application/json" };
+	if (apiKey) {
+		headers.Authorization = `Bearer ${apiKey}`;
+	}
+	return fetch(MODELS_URL, { headers, signal });
+}
+
+function isInputModality(value: string): value is "text" | "image" {
+	return value === "text" || value === "image";
+}
+
+/** Resolve the input modalities the catalog reports, falling back to curated metadata. */
+function resolveInput(entry: IncoCatalogEntry, curated: CuratedModel | undefined): ("text" | "image")[] {
+	const reported = entry.modalities?.input?.filter(isInputModality) ?? [];
+	if (reported.length === 0) {
+		return curated?.input ?? ["text"];
+	}
+	return reported.includes("text") ? reported : (["text", ...reported] as ("text" | "image")[]);
+}
+
+/** Convert one Inco catalog entry into a Pi chat model. */
+function toModel(entry: IncoCatalogEntry): Model<"openai-completions"> {
+	const curated = CURATED[entry.id];
+	const reasoningEffort = entry.capabilities?.reasoning_effort === true;
+	return {
+		id: entry.id,
+		name: entry.name || entry.id,
+		api: "openai-completions",
+		provider: PROVIDER_ID,
+		baseUrl: BASE_URL,
+		reasoning: true,
+		input: resolveInput(entry, curated),
+		cost: {
+			input: entry.pricing?.input ?? 0,
+			output: entry.pricing?.output ?? 0,
+			cacheRead: entry.pricing?.cached_input ?? 0,
+			cacheWrite: 0,
+		},
+		contextWindow: entry.context_length ?? DEFAULT_CONTEXT_WINDOW,
+		maxTokens: curated?.maxTokens ?? DEFAULT_MAX_TOKENS,
+		compat: {
+			// Inco is not OpenAI: send only fields its OpenAI-compatible surface documents.
+			supportsStore: false,
+			supportsDeveloperRole: false,
+			supportsReasoningEffort: reasoningEffort,
+			supportsUsageInStreaming: true,
+			maxTokensField: "max_completion_tokens",
+			// Inco returns reasoning in `reasoning_content` and expects it replayed.
+			requiresReasoningContentOnAssistantMessages: true,
+			thinkingFormat: "openai",
+		},
+	};
+}
+
+/** Fetch the live catalog, authenticating when a key is available. */
+async function fetchIncoModels(context: RefreshModelsContext): Promise<Model<"openai-completions">[]> {
+	const apiKey = context.credential?.type === "oauth" ? context.credential.access : context.credential?.key;
+	const response = await requestCatalog(apiKey, context.signal);
+	if (!response.ok) {
+		throw await catalogError(response);
+	}
+
+	const payload = (await response.json()) as { data?: IncoCatalogEntry[] };
+	const entries = Array.isArray(payload?.data) ? payload.data : [];
+	if (entries.length === 0) {
+		throw new Error("Inco model catalog returned no models");
+	}
+
+	return entries.filter((entry) => entry && typeof entry.id === "string").map(toModel);
+}
+
+/**
+ * Standard api-key auth plus a login that verifies the key against the catalog
+ * before Pi persists it, so a typo fails in the login dialog instead of on the
+ * first message.
+ */
+function incoApiKeyAuth(): ApiKeyAuth {
+	const standard = envApiKeyAuth("Inco AI API key", ["INCO_API_KEY"]);
+	return {
+		...standard,
+		async login(interaction) {
+			if (!standard.login) {
+				throw new Error("Inco AI API key login is unavailable");
+			}
+			const credential = await standard.login(interaction);
+			const key = credential.key?.trim();
+			if (!key) {
+				throw new Error("No API key provided");
+			}
+			interaction.notify({ type: "progress", message: "Verifying API key…" });
+			const response = await requestCatalog(key, interaction.signal);
+			if (!response.ok) {
+				throw await catalogError(response);
+			}
+			await response.arrayBuffer();
+			return { ...credential, key };
+		},
+	};
+}
+
+export default function incoProviderExtension(pi: ExtensionAPI): void {
+	pi.registerProvider(
+		createProvider({
+			id: PROVIDER_ID,
+			name: PROVIDER_NAME,
+			baseUrl: BASE_URL,
+			auth: { apiKey: incoApiKeyAuth() },
+			models: CATALOG_SNAPSHOT.map(toModel),
+			api: openAICompletionsApi(),
+			fetchModels: fetchIncoModels,
+		}),
+	);
+}

package/package.json

@@ -0,0 +1,37 @@
+{
+	"name": "pi-incoai",
+	"version": "0.1.0",
+	"description": "Inco AI (inco.ai) model provider extension for the Pi coding agent",
+	"keywords": [
+		"pi-package",
+		"pi",
+		"pi-coding-agent",
+		"inco",
+		"incoai",
+		"inference",
+		"provider"
+	],
+	"license": "MIT",
+	"author": "Sergiu Truta",
+	"homepage": "https://github.com/serggiu/pi-incoai#readme",
+	"bugs": {
+		"url": "https://github.com/serggiu/pi-incoai/issues"
+	},
+	"type": "module",
+	"files": [
+		"extensions",
+		"README.md",
+		"LICENSE"
+	],
+	"repository": {
+		"type": "git",
+		"url": "git+https://github.com/serggiu/pi-incoai.git"
+	},
+	"publishConfig": {
+		"access": "public"
+	},
+	"peerDependencies": {
+		"@earendil-works/pi-ai": "*",
+		"@earendil-works/pi-coding-agent": "*"
+	}
+}