decorated-pi 0.1.0 → 0.2.1

package/README.md

@@ -1,70 +1,46 @@
 # decorated-pi
 
-`decorated-pi` is a Pi extension bundle that adds safety gates, LSP tools, image/compaction model helpers, smarter `@` file completion, dynamic subdirectory `AGENTS.md` loading, and a few workflow quality-of-life improvements.
-
-## Status
-
-Current scope is **functionally complete for local use**.
-
-Recent audit highlights:
-- Fixed stale LSP diagnostics caused by returning cached diagnostics after `didChange`
-- Fixed `subdir-agents` path capture on `tool_call`
-- Fixed a `smart-at` multi-token search bug
-- Updated safety checks so shell overwrite attempts on existing files are treated as dangerous
+`decorated-pi` is a Pi extension that adds safety gates, LSP tools, image/compaction model helpers, smarter `@` file search, dynamic subdirectory `AGENTS.md` loading, and a few workflow quality-of-life improvements.
 
 ## Features
 
-### 1. Decorated Pi Guidance
-Adds global system-prompt guidance via `before_agent_start.systemPrompt`:
-- Restate understanding before acting
-- Break medium/large tasks into discrete steps
-
-### 2. Safety Layer
-Implemented in `extensions/safety.ts`.
+### 1. Safety Layer
 
 - **Dangerous bash guard**
-  - Blocks or asks for confirmation on destructive commands such as:
+  - asks for confirmation on destructive commands such as:
     - `rm`
     - `sudo`
-    - `svn commit`
-    - `svn revert`
-    - `git reset`
-    - `git restore`
-    - `git clean`
-    - `git push`
-    - `git revert`
-- **Shell overwrite detection**
-  - Detects `bash` commands that would overwrite an **existing regular file**, including:
-    - `>` / `>>`
-    - `1>` / `1>>`
-    - `2>` / `2>>`
-    - `&>` / `&>>`
-    - `tee`
-  - Aggregates **all** dangerous reasons found in one command
+    - `svn commit/revert`
+    - `git reset/restore/clean/push/revert`
+    - `npm publish`
+    - `>` / `1>` / `2>` / `&>` / `tee` overwrite existing files
+  - Hints the agent to use `edit` instead of `write` on non-empty files
 - **Protected paths**
-  - Blocks `write` / `edit` to sensitive locations such as `.env`, `.git/`, `.ssh/`, `node_modules/`, `*.pem`, `*.key`, etc.
-- **Write guard**
-  - Blocks the `write` tool when it would overwrite a non-empty file
-  - Instructs the model to use `edit` instead
+  - Blocks read/write access (via `write`/`edit`/`read` tools or `cat`/`head`/`tail`/`grep`/`rg` etc. bash commands) to sensitive locations such as `.env`, `.git/`, `.ssh/`, `*.pem`, `*.key`, etc.
 - **Secret redaction**
-  - Uses `secretlint` rules to redact secrets from tool output before they are fed back into context
+  - Dual-layer detection: 40+ known-format patterns (AWS, GitHub, OpenAI, etc.) + Adjusted Shannon Entropy analysis for unknown formats. Based on [opencode-secrets-protect](https://github.com/jscheel/opencode-secrets-protect) (MIT)
+
+### 2. Smart `@` File Search
 
-### 3. Smart `@` Completion
-Implemented in `extensions/smart-at.ts`.
+Replaces Pi's default file search with a faster project-aware search strategy:
 
-Replaces Pi's default file autocomplete behavior with a faster project-aware search strategy:
 - Uses `git ls-files` in git repos
 - Falls back to `fd` outside git repos
 - Caches results for 10 seconds
 - Scores primarily on **filename match quality**, not full-path fuzziness
 - Penalizes hidden/cache/build directories
 - Hides hidden paths from empty-query results
-- Keeps Pi's original `applyCompletion` / `shouldTriggerFileCompletion` binding behavior intact
 
-### 4. LSP Tool Suite
-Implemented in `extensions/lsp/`.
+### 3. LSP Tool Suite
+
+Based on [@spences10/pi-lsp](https://github.com/spences10/my-pi/tree/main/packages/pi-lsp) by Scott Spence (MIT License), with additions:
+
+- C/C++ (clangd) and Lua support
+- `lsp_find_symbol`, `lsp_rename`, multi-file support merged into `lsp_diagnostics`
+- Force-sync on `didChange` (no stale diagnostics)
 
 Registered tools:
+
 - `lsp_diagnostics`
 - `lsp_find_symbol`
 - `lsp_hover`
@@ -74,8 +50,8 @@ Registered tools:
 - `lsp_rename`
 
 Supported languages:
-- c
-- cpp
+
+- c/cpp
 - go
 - java
 - lua
@@ -85,134 +61,49 @@ Supported languages:
 - svelte
 - typescript
 
-LSP integration includes:
-- prompt snippets for `Available tools`
-- tool-specific prompt guidelines
-- parameter descriptions for the JSON schema
-- trust checks for project-local LSP binaries
-- an LSP-specific system-prompt section injected only when LSP tools are active
+### 4. Auxiliary Models (Image + Compact)
 
-### 5. Image Read Fallback
-Implemented in `extensions/extend-model.ts`.
+Uses cheaper models for auxiliary tasks, configured via `/extend-model`:
 
-When the model reads an image file and an image model is configured:
-- Detects supported image types via magic bytes
-- Calls a configured vision-capable model
-- Replaces the read result with image analysis text
+- **Image read fallback** — when the model reads an image file, detects type via magic bytes, calls a configured vision-capable model, and replaces the read result with image analysis text (jpeg, png, gif, webp)
+- **Compact model** — uses a configured model for context compaction (instead of the main model), auto-resumes after compaction.
 
-Supported image types:
-- jpeg
-- png
-- gif
-- webp
-
-### 6. Custom Compact Model + Auto Resume
-Also implemented in `extensions/extend-model.ts`.
-
-- Supports a configured **compact model** through `session_before_compact`
-- Preserves auto-resume behavior through `session_compact`
-- Appends read/modified file summaries to compaction output
-
-### 7. `/extend-model`
-Implemented in `extensions/slash.ts`.
-
-Interactive command for configuring:
-- image model
-- compact model
-
-### 8. `/retry`
-Implemented in `extensions/slash.ts`.
-
-Allows continuing after interruption by:
-- aborting the current run if needed
-- sending a hidden continuation trigger
-- injecting a one-turn retry note into the system prompt
-
-### 9. Dynamic Subdirectory `AGENTS.md` / `CLAUDE.md`
-Implemented in `extensions/subdir-agents.ts`.
+### 5. Dynamic Subdirectory `AGENTS.md` / `CLAUDE.md`
 
 When the agent reads or edits a file:
+
 - discovers `AGENTS.md` / `CLAUDE.md` in the file's directory and ancestor directories
 - injects newly discovered guidance into tool results
 - persists discovered files into the session so they are restored on resume
 
-### 10. Automatic Session Title
-Implemented in `extensions/session-title.ts`.
-
-- Derives the session name from the first user message
-- Avoids overriding a manually assigned session name
+### 6. Extend Providers
 
-## Install
+Extend providers are registered via `/login` → "Use a subscription":
 
-### Local install
+| Provider | Base URL |
+| ---------- | ----------- |
+| Ollama Cloud | `ollama.com/v1` |
+| Baidu Qianfan | `qianfan.baidubce.com/v2/coding` |
+| ARK Coding | `ark.cn-beijing.volces.com/api/coding/v3` |
 
-```bash
-pi install /path/to/decorated-pi
-```
-
-Then reload Pi:
+## Install
 
 ```bash
-/reload
+pi install /path/to/decorated-pi #local
+pi install npm:decorated-pi #npm
+pi install git:github.com/lcwecker/decorated-pi #github
 ```
 
-### npm publish
-Not published yet.
+Then reload Pi
 
 ## Configuration
 
 Runtime settings are stored in:
 
 ```text
-~/.pi/agent/extensions/decorated-pi.json
-```
-
-Current keys:
-- `imageModelKey`
-- `compactModelKey`
-
-Design rule:
-- **only** `extensions/settings.ts` writes this file
-- all other modules read through exported getters/setters
-
-## Architecture
-
-Entry point:
-- `extensions/index.ts`
-
-Main modules:
-- `extensions/guidance.ts` — global system prompt guidance
-- `extensions/safety.ts` — command guard, protected paths, write guard, secret redaction
-- `extensions/smart-at.ts` — smart `@` autocomplete
-- `extensions/extend-model.ts` — image fallback, compact model, auto-resume
-- `extensions/slash.ts` — `/extend-model`, `/retry`
-- `extensions/subdir-agents.ts` — dynamic `AGENTS.md` / `CLAUDE.md`
-- `extensions/session-title.ts` — session naming
-- `extensions/lsp/*` — LSP client, server manager, prompt wiring, tools
-- `extensions/settings.ts` — config I/O
-
-## Current Limitations
-
-- `write` protection applies to the **LLM's `write` tool**; agent attempts to overwrite files via `bash` are handled separately by the dangerous-command gate
-- shell overwrite detection only escalates when the target resolves to an **existing regular file**
-- compact behavior still relies on Pi's normal compaction flow; this extension customizes the model and post-compact continuation, not the global compaction threshold UI
-- npm distribution is not set up yet
-- there is no dedicated automated test suite yet; current validation is based on Pi runtime checks, LSP diagnostics, and manual smoke tests
-
-## Development Notes
-
-Helpful checks during development:
-
-```bash
-pi install /path/to/decorated-pi
-/reload
+~/.pi/agent/decorated-pi.json
 ```
 
-For focused validation, use:
-- `lsp_diagnostics` on edited source files
-- real `pi "..."` smoke tests for safety behavior
-- manual autocomplete checks for `smart-at`
-
 ## License
 
 MIT

package/extensions/extend-model.ts

@@ -387,12 +387,7 @@ export function setupExtendModel(pi: ExtensionAPI) {
           auth.apiKey ?? "", auth.headers, signal, customInstructions, previousSummary);
       }
 
-      const readFiles = [...fileOps.read].filter(f => !fileOps.edited.has(f) && !fileOps.written.has(f)).sort();
-      const modifiedFiles = [...new Set([...fileOps.edited, ...fileOps.written])].sort();
-      if (readFiles.length > 0) summary += `\n\n<read-files>\n${readFiles.join("\n")}\n</read-files>`;
-      if (modifiedFiles.length > 0) summary += `\n\n<modified-files>\n${modifiedFiles.join("\n")}\n</modified-files>`;
-
-      return { compaction: { summary, firstKeptEntryId, tokensBefore, details: { readFiles, modifiedFiles } } };
+      return { compaction: { summary, firstKeptEntryId, tokensBefore } };
     } catch (err) {
       if (signal.aborted) return;
       ctx.ui.notify(`Compact failed: ${err instanceof Error ? err.message : err}`, "error");

package/extensions/index.ts

@@ -10,6 +10,7 @@ import { setupSubdirAgents } from "./subdir-agents";
 import { setupSessionTitle } from "./session-title";
 import { setupGuidance } from "./guidance";
 import { setupLsp } from "./lsp/index";
+import { setupProviders } from "./providers/index";
 import { setupSmartAt } from "./smart-at";
 
 export default function (pi: ExtensionAPI) {
@@ -20,5 +21,6 @@ export default function (pi: ExtensionAPI) {
   setupSessionTitle(pi);
   setupGuidance(pi);
   setupLsp(pi);
+  setupProviders(pi);
   setupSmartAt(pi);
 }

package/extensions/lsp/client.ts

@@ -1,3 +1,11 @@
+/**
+ * LSP Client — JSON-RPC over stdio transport
+ *
+ * Based on @spences10/pi-lsp by Scott Spence
+ * https://github.com/spences10/my-pi/tree/main/packages/pi-lsp (MIT License)
+ *
+ * Modifications: added rename() method, simplified type exports
+ */
 import { spawn, ChildProcess } from "node:child_process";
 import { EventEmitter } from "node:events";
 import { pathToFileURL } from "node:url";
@@ -204,7 +212,10 @@ export class LspClient extends EventEmitter {
   async ensure_document_open(uri: string, text: string): Promise<void> {
     const existing = this.#open_docs.get(uri);
     if (existing) {
-      if (existing.text === text) return;
+      // Always force-sync: the file may have been modified externally
+      // (e.g. by pi's edit/write tools), and even if text matches the
+      // cached version, diagnostics may be stale due to changes in
+      // other files. Never skip the sync based on text comparison.
       const next_version = existing.version + 1;
       this.#open_docs.set(uri, { version: next_version, text });
       this.#diagnostics_by_uri.delete(uri);

package/extensions/lsp/env.ts

@@ -1,3 +1,9 @@
+/**
+ * LSP Child Process Environment — restricted env for LSP server processes
+ *
+ * Based on @spences10/pi-lsp by Scott Spence
+ * https://github.com/spences10/my-pi/tree/main/packages/pi-lsp (MIT License)
+ */
 import { create_child_process_env as create_shared_child_process_env } from "@spences10/pi-child-env";
 
 export function create_child_process_env(

package/extensions/lsp/format.ts

@@ -1,3 +1,9 @@
+/**
+ * LSP Output Formatting — diagnostics, hover, locations, symbols
+ *
+ * Based on @spences10/pi-lsp by Scott Spence
+ * https://github.com/spences10/my-pi/tree/main/packages/pi-lsp (MIT License)
+ */
 import { fileURLToPath } from "node:url";
 import type { LspDiagnostic, LspHover, LspLocation, LspDocumentSymbol } from "./client.js";
 import { LspClientStartError } from "./client.js";

package/extensions/lsp/index.ts

@@ -1,3 +1,9 @@
+/**
+ * LSP Extension Entry Point
+ *
+ * Based on @spences10/pi-lsp by Scott Spence
+ * https://github.com/spences10/my-pi/tree/main/packages/pi-lsp (MIT License)
+ */
 import { LspServerManager } from "./server-manager.js";
 import { register_lsp_tools } from "./tools.js";
 import { setup_lsp_prompt } from "./prompt.js";

package/extensions/lsp/prompt.ts

@@ -1,3 +1,9 @@
+/**
+ * LSP System Prompt — injects LSP guidance into agent system prompt
+ *
+ * Based on @spences10/pi-lsp by Scott Spence
+ * https://github.com/spences10/my-pi/tree/main/packages/pi-lsp (MIT License)
+ */
 import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
 import { list_supported_languages } from "./servers.js";
 

package/extensions/lsp/server-manager.ts

@@ -1,3 +1,9 @@
+/**
+ * LSP Server Manager — lifecycle, per-language instances, project-local trust
+ *
+ * Based on @spences10/pi-lsp by Scott Spence
+ * https://github.com/spences10/my-pi/tree/main/packages/pi-lsp (MIT License)
+ */
 import { resolve_project_trust } from "@spences10/pi-project-trust";
 import { readFile } from "node:fs/promises";
 import { isAbsolute, resolve } from "node:path";

package/extensions/lsp/servers.ts

@@ -1,3 +1,11 @@
+/**
+ * LSP Server Config — language detection, server commands, workspace roots
+ *
+ * Based on @spences10/pi-lsp by Scott Spence
+ * https://github.com/spences10/my-pi/tree/main/packages/pi-lsp (MIT License)
+ *
+ * Modifications: added C/C++ (clangd) and Lua support
+ */
 import { existsSync } from "node:fs";
 import { dirname, extname, isAbsolute, join, resolve } from "node:path";
 
@@ -31,7 +39,7 @@ const EXTENSION_LANGUAGES: Record<string, string> = {
   ".svelte": "svelte",
 };
 
-type LanguageConfig = {
+export type LanguageConfig = {
   language: string;
   command: string;
   args: string[];

package/extensions/lsp/tools.ts

@@ -1,3 +1,11 @@
+/**
+ * LSP Tool Definitions — diagnostics, hover, definition, references, symbols, rename
+ *
+ * Based on @spences10/pi-lsp by Scott Spence
+ * https://github.com/spences10/my-pi/tree/main/packages/pi-lsp (MIT License)
+ *
+ * Modifications: added lsp_find_symbol, lsp_rename, multi-file lsp_diagnostics
+ */
 import { defineTool, type ExtensionAPI } from "@earendil-works/pi-coding-agent";
 import { Type } from "typebox";
 import { list_supported_languages } from "./servers.js";

package/extensions/lsp/trust.ts

@@ -1,3 +1,9 @@
+/**
+ * LSP Binary Trust Store — project-local binary trust management
+ *
+ * Based on @spences10/pi-lsp by Scott Spence
+ * https://github.com/spences10/my-pi/tree/main/packages/pi-lsp (MIT License)
+ */
 import { getAgentDir } from "@earendil-works/pi-coding-agent";
 import {
   is_project_subject_trusted,

package/extensions/providers/ark-coding.ts

@@ -0,0 +1,73 @@
+/**
+ * ARK Coding Plan — OAuth/subscription provider with hardcoded models
+ *
+ * Provider: "ark-coding"
+ * Base URL: https://ark.cn-beijing.volces.com/api/coding/v3 (OpenAI-compatible)
+ * Auth: OAuth/subscription login → prompt for API key
+ *
+ * All models hardcoded. No dynamic fetching, no config file caching.
+ * - No auth → no models in /model (clean UX, via hasConfiguredAuth)
+ * - Login → models become available immediately
+ * - Startup → models registered unconditionally (hardcoded)
+ */
+
+import type { ExtensionAPI, ProviderModelConfig } from "@earendil-works/pi-coding-agent";
+import type { OAuthCredentials, OAuthLoginCallbacks } from "@earendil-works/pi-ai";
+
+const PROVIDER_ID = "ark-coding";
+const PROVIDER_DISPLAY_NAME = "ARK Coding Plan";
+const BASE_URL = "https://ark.cn-beijing.volces.com/api/coding/v3";
+
+// ── Hardcoded models (parameters from models.dev) ─────────────────────────
+
+const MODELS: ProviderModelConfig[] = [
+  { id: "deepseek-v3.2", name: "DeepSeek V3.2", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 163_840, maxTokens: 65_536, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "glm-4.7", name: "GLM 4.7", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 202_752, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "glm-5.1", name: "GLM 5.1", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 202_752, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "kimi-k2.5", name: "Kimi K2.5", reasoning: true, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 262_144, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "kimi-k2.6", name: "Kimi K2.6", reasoning: true, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 262_144, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "minimax-m2.5", name: "MiniMax M2.5", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 204_800, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "minimax-m2.7", name: "MiniMax M2.7", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 196_608, maxTokens: 196_608, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "doubao-seed-2.0-code", name: "Doubao Seed 2.0 Code", reasoning: false, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 256_000, maxTokens: 128_000, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "doubao-seed-2.0-pro", name: "Doubao Seed 2.0 Pro", reasoning: false, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 256_000, maxTokens: 128_000, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "doubao-seed-2.0-lite", name: "Doubao Seed 2.0 Lite", reasoning: false, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 256_000, maxTokens: 32_000, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "doubao-seed-code", name: "Doubao Seed Code", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 256_000, maxTokens: 16_384, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+];
+
+// ── Entry ──────────────────────────────────────────────────────────────────
+
+export function setupArkCoding(pi: ExtensionAPI) {
+  pi.registerProvider(PROVIDER_ID, {
+    name: PROVIDER_DISPLAY_NAME,
+    baseUrl: BASE_URL,
+    api: "openai-completions",
+    authHeader: true,
+    models: MODELS,
+    oauth: {
+      name: PROVIDER_DISPLAY_NAME,
+
+      async login(callbacks: OAuthLoginCallbacks): Promise<OAuthCredentials> {
+        const apiKey = (await callbacks.onPrompt({
+          message: "Enter ARK Coding Plan API key:",
+          placeholder: "your-api-key",
+        })).trim();
+
+        if (!apiKey) throw new Error("API key cannot be empty.");
+
+        return {
+          refresh: apiKey,
+          access: apiKey,
+          expires: Date.now() + 1000 * 365.24 * 24 * 3600 * 1000, // ~1000 years
+        };
+      },
+
+      refreshToken(cred: OAuthCredentials): Promise<OAuthCredentials> {
+        return Promise.resolve(cred); // API key never expires
+      },
+
+      getApiKey(cred: OAuthCredentials): string {
+        return cred.access;
+      },
+    },
+  });
+}

package/extensions/providers/index.ts

@@ -0,0 +1,9 @@
+import { setupArkCoding } from "./ark-coding.js";
+import { setupOllamaCloud } from "./ollama-cloud.js";
+import { setupQianfanCoding } from "./qianfan-coding.js";
+
+export function setupProviders(pi: any) {
+  setupArkCoding(pi);
+  setupOllamaCloud(pi);
+  setupQianfanCoding(pi);
+}

package/extensions/providers/ollama-cloud.ts

@@ -0,0 +1,101 @@
+/**
+ * Ollama Cloud — OAuth/subscription provider with hardcoded models
+ *
+ * Provider: "ollama-cloud"
+ * Base URL: https://ollama.com/v1 (OpenAI-compatible)
+ * Auth: OAuth/subscription login → prompt for API key
+ *
+ * All models hardcoded from models.dev. No dynamic fetching, no config file caching.
+ * - No auth → no models in /model (clean UX, via hasConfiguredAuth)
+ * - Login → models become available immediately
+ * - Startup → models registered unconditionally (hardcoded)
+ */
+
+import type { ExtensionAPI, ProviderModelConfig } from "@earendil-works/pi-coding-agent";
+import type { OAuthCredentials, OAuthLoginCallbacks } from "@earendil-works/pi-ai";
+
+const PROVIDER_ID = "ollama-cloud";
+const PROVIDER_DISPLAY_NAME = "Ollama Cloud";
+const BASE_URL = "https://ollama.com/v1";
+
+// ── Hardcoded models (from models.dev) ────────────────────────────────────
+
+const MODELS: ProviderModelConfig[] = [
+  { id: "cogito-2.1:671b", name: "cogito-2.1:671b", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 163_840, maxTokens: 32_000, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "deepseek-v3.1:671b", name: "deepseek-v3.1:671b", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 163_840, maxTokens: 163_840, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "deepseek-v3.2", name: "deepseek-v3.2", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 163_840, maxTokens: 65_536, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "deepseek-v4-flash", name: "deepseek-v4-flash", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 1_048_576, maxTokens: 1_048_576, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "deepseek-v4-pro", name: "deepseek-v4-pro", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 1_048_576, maxTokens: 1_048_576, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "devstral-2:123b", name: "devstral-2:123b", reasoning: false, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 262_144, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "devstral-small-2:24b", name: "devstral-small-2:24b", reasoning: false, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 262_144, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "gemini-3-flash-preview", name: "gemini-3-flash-preview", reasoning: true, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 1_048_576, maxTokens: 65_536, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "gemma3:12b", name: "gemma3:12b", reasoning: false, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 131_072, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "gemma3:27b", name: "gemma3:27b", reasoning: false, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 131_072, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "gemma3:4b", name: "gemma3:4b", reasoning: false, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 131_072, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "gemma4:31b", name: "gemma4:31b", reasoning: true, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 262_144, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "glm-4.6", name: "glm-4.6", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 202_752, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "glm-4.7", name: "glm-4.7", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 202_752, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "glm-5", name: "glm-5", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 202_752, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "glm-5.1", name: "glm-5.1", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 202_752, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "gpt-oss:120b", name: "gpt-oss:120b", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 131_072, maxTokens: 32_768, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "gpt-oss:20b", name: "gpt-oss:20b", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 131_072, maxTokens: 32_768, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "kimi-k2-thinking", name: "kimi-k2-thinking", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 262_144, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "kimi-k2.5", name: "kimi-k2.5", reasoning: true, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 262_144, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "kimi-k2.6", name: "kimi-k2.6", reasoning: true, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 262_144, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "kimi-k2:1t", name: "kimi-k2:1t", reasoning: false, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 262_144, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "minimax-m2", name: "minimax-m2", reasoning: false, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 204_800, maxTokens: 128_000, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "minimax-m2.1", name: "minimax-m2.1", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 204_800, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "minimax-m2.5", name: "minimax-m2.5", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 204_800, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "minimax-m2.7", name: "minimax-m2.7", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 196_608, maxTokens: 196_608, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "ministral-3:14b", name: "ministral-3:14b", reasoning: false, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 128_000, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "ministral-3:3b", name: "ministral-3:3b", reasoning: false, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 128_000, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "ministral-3:8b", name: "ministral-3:8b", reasoning: false, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 128_000, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "mistral-large-3:675b", name: "mistral-large-3:675b", reasoning: false, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 262_144, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "nemotron-3-nano:30b", name: "nemotron-3-nano:30b", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 1_048_576, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "nemotron-3-super", name: "nemotron-3-super", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 65_536, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "qwen3-coder-next", name: "qwen3-coder-next", reasoning: false, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 65_536, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "qwen3-coder:480b", name: "qwen3-coder:480b", reasoning: false, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 65_536, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "qwen3-next:80b", name: "qwen3-next:80b", reasoning: true, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 32_768, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "qwen3-vl:235b", name: "qwen3-vl:235b", reasoning: true, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 32_768, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "qwen3-vl:235b-instruct", name: "qwen3-vl:235b-instruct", reasoning: false, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 131_072, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "qwen3.5:397b", name: "qwen3.5:397b", reasoning: true, input: ["text", "image"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 262_144, maxTokens: 65_536, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+  { id: "rnj-1:8b", name: "rnj-1:8b", reasoning: false, input: ["text"], cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: 32_768, maxTokens: 4_096, compat: { supportsDeveloperRole: false, supportsReasoningEffort: true } as any },
+];
+
+// ── Entry ──────────────────────────────────────────────────────────────────
+
+export function setupOllamaCloud(pi: ExtensionAPI) {
+  pi.registerProvider(PROVIDER_ID, {
+    name: PROVIDER_DISPLAY_NAME,
+    baseUrl: BASE_URL,
+    api: "openai-completions",
+    authHeader: true,
+    models: MODELS,
+    oauth: {
+      name: PROVIDER_DISPLAY_NAME,
+
+      async login(callbacks: OAuthLoginCallbacks): Promise<OAuthCredentials> {
+        const apiKey = (await callbacks.onPrompt({
+          message: "Enter Ollama Cloud API key:",
+          placeholder: "your-api-key",
+        })).trim();
+
+        if (!apiKey) throw new Error("API key cannot be empty.");
+
+        return {
+          refresh: apiKey,
+          access: apiKey,
+          expires: Date.now() + 1000 * 365.24 * 24 * 3600 * 1000, // ~1000 years
+        };
+      },
+
+      refreshToken(cred: OAuthCredentials): Promise<OAuthCredentials> {
+        return Promise.resolve(cred); // API key never expires
+      },
+
+      getApiKey(cred: OAuthCredentials): string {
+        return cred.access;
+      },
+    },
+  });
+}