pi-wafer-provider 1.0.2

package/.github/FUNDING.yml

@@ -0,0 +1,4 @@
+github: monotykamary
+ko_fi: monotykamary
+buy_me_a_coffee: monotykamary
+polar: monotykamary

package/.pi/autoresearch/session-id

@@ -0,0 +1 @@
+019da9d5-6cb9-764d-a39e-ce1dbd18a3b2

package/.pi/messenger/channels/memory.jsonl

@@ -0,0 +1 @@
+{"_meta":true,"v":1,"id":"memory","type":"named","createdAt":"2026-05-22T02:22:08.562Z","description":"Cross-session knowledge and insights"}

package/.pi/messenger/session-id

@@ -0,0 +1 @@
+019e9c1b-cad3-7bd6-a910-c015d0788505

package/AGENTS.md

@@ -0,0 +1,56 @@
+# AGENTS.md
+
+## DO NOT EDIT — Auto-generated Files
+
+The following files are **idempotent** and regenerated by `scripts/update-models.js`. Never edit them directly — your changes will be overwritten on the next model sync.
+
+| File | Why it's auto-generated |
+|------|------------------------|
+| `models.json` | Built from the provider API. `update-models.js` fetches models, preserves curated data for known IDs, and writes this file. |
+| `README.md` (model table) | The table under `## Available Models` is replaced in-place by `update-models.js` after merging base models → patch → custom models. |
+
+## Correct Files to Edit
+
+When a model needs overrides, new properties, or corrections, edit the appropriate source file below. These are the **source of truth** that the update script reads but never writes.
+
+| File | Purpose |
+|------|---------|
+| `patch.json` | Per-model overrides keyed by model ID. Add reasoning flags, compat settings, pricing corrections, thinking level maps, etc. Applied on top of `models.json` at runtime and for README generation. |
+| `custom-models.json` | Models that don't exist in the provider API (hidden models, router endpoints, cross-provider aliases). Merged after patch. Format: array of full model objects (same schema as `models.json` entries). |
+| `index.ts` | Provider extension code. |
+| `scripts/update-models.js` | The sync script itself (edit only if changing how models are fetched/transformed). |
+
+## Data Flow
+
+```
+Provider API  ──fetch──►  models.json  ──apply──►  patch.json  ──merge──►  custom-models.json
+                                    │                            │                      │
+                                    └────────────────────────────┴──────────────────────┘
+                                                      │
+                                              README model table
+```
+
+1. `models.json` — base data from the provider API (auto-generated, DO NOT EDIT)
+2. `patch.json` — overrides applied on top (EDIT THIS for corrections/enrichments)
+3. `custom-models.json` — additional models not in the API (EDIT THIS for new models)
+4. README table — rendered from the merged result of all three (auto-generated, DO NOT EDIT)
+
+## Common Tasks
+
+### Add a compat setting or override pricing for an existing model
+→ Edit `patch.json`. Add an entry keyed by the model's `id`.
+
+### Add a model not available in the provider API
+→ Edit `custom-models.json`. Add a full model object to the array.
+
+### Update models from the provider API
+→ Run `node scripts/update-models.js` (may require an API key env var).
+
+### Regenerate the README model table
+→ Run `node scripts/update-models.js` — it updates both `models.json` and the README table.
+
+## TL;DR
+
+- **Never edit `models.json`** — edit `patch.json` instead.
+- **Never edit the README model table** — run the update script instead.
+- `patch.json` and `custom-models.json` are the source files you should modify.

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026
+
+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 +1,137 @@
+<div align="center">
+
+# 🧇 pi-wafer-provider
+
+**Qwen3.5, GLM-5.1, Kimi K2.6 & more via [Wafer](https://wafer.ai)**
+
+_Serverless inference with free DeepSeek V4 models for [pi](https://github.com/earendil-works/pi-coding-agent)._
+
+[![pi extension](https://img.shields.io/badge/pi-extension-blueviolet)](https://github.com/earendil-works/pi-coding-agent)
+[![license](https://img.shields.io/badge/license-MIT-blue)](./LICENSE)
+
+</div>
+
+---
+
+## Features
+
+- **Fast Open-Source Models** via Wafer Serverless
+- **Unified API** via Wafer's OpenAI-compatible completions endpoint
+- **Cost Tracking** with per-model pricing for budget management
+- **Reasoning Models** support for advanced reasoning capabilities
+- **Vision Support** for Qwen3.5 (image + text input)
+
+## Provider
+
+| Provider | API Key Env Var | Auth JSON Key |
+|----------|----------------|---------------|
+| `wafer-serverless` | `WAFER_SERVERLESS_API_KEY` | `wafer-serverless` |
+
+Use `/model` in pi to select from available models.
+
+## Installation
+
+### Option 1: Using `pi install` (Recommended)
+
+Install directly from GitHub:
+
+```bash
+pi install https://github.com/monotykamary/pi-wafer-provider
+```
+
+Then set your API key and run pi:
+```bash
+# Recommended: add to auth.json
+# See Authentication section below
+
+# Or set as environment variables
+export WAFER_SERVERLESS_API_KEY=your-serverless-key
+
+pi
+```
+
+### Option 2: Manual Clone
+
+1. Clone this repository:
+   ```bash
+   git clone https://github.com/monotykamary/pi-wafer-provider.git
+   cd pi-wafer-provider
+   ```
+
+2. Set your Wafer API key:
+   ```bash
+   # Recommended: add to auth.json
+   # See Authentication section below
+
+   # Or set as environment variables
+   export WAFER_SERVERLESS_API_KEY=your-serverless-key
+   ```
+
+3. Run pi with the extension:
+   ```bash
+   pi -e /path/to/pi-wafer-provider
+   ```
+
+## Available Models
+
+| Model | Type | Context | Max Output | Input Cost | Output Cost | Cached Input |
+|-------|------|---------|------------|------------|-------------|--------------|
+| Deepseek V4 Flash | Text | 1M | 16K | Free | Free | Free |
+| Deepseek V4 Pro | Text | 1M | 384K | Free | Free | Free |
+| GLM 5.1 | Text | 203K | 33K | $1.50 | $4.50 | $0.15 |
+| GLM 5.2 | Text | 1M | 16K | Free | Free | Free |
+| Kimi K2.6 | Text | 262K | 33K | $1.10 | $4.80 | $0.11 |
+| Kimi K2.7 Code | Text | 262K | 16K | Free | Free | Free |
+| MiniMax M3 | Text | 1M | 16K | Free | Free | Free |
+| Qwen 3.5 397B (A17B) | Text + Image | 262K | 33K | $0.60 | $3.60 | $0.06 |
+| Qwen3.6 35B A3B | Text | 256K | 16K | Free | Free | Free |
+| Qwen3.7 Max | Text | 256K | 16K | $5.00 | $15.00 | $0.50 |
+
+*Costs are per million tokens. Prices based on official provider pricing.*
+
+## Usage
+
+After loading the extension, use the `/model` command in pi to select your preferred model:
+
+```
+/model
+```
+
+Then select `wafer-serverless` as the provider and choose from the available models.
+
+## Authentication
+
+API keys can be configured in multiple ways (resolved in this order):
+
+1. **`auth.json`** (recommended) — Add to `~/.pi/agent/auth.json`:
+   ```json
+   {
+     "wafer-serverless":  { "type": "api_key", "key": "your-serverless-key" }
+   }
+   ```
+   The `key` field supports literal values, env var names, and shell commands (prefix with `!`). See [pi's auth file docs](https://github.com/badlogic/pi-mono) for details.
+2. **Runtime override** — Use the `--api-key` CLI flag
+3. **Environment variable** — Set `WAFER_SERVERLESS_API_KEY` (falls back to `WAFER_API_KEY` for backwards compatibility)
+
+## Environment Variables
+
+| Variable | Provider | Description |
+|----------|----------|-------------|
+| `WAFER_SERVERLESS_API_KEY` | `wafer-serverless` | Primary API key |
+| `WAFER_API_KEY` | `wafer-serverless` | Legacy fallback (used if `WAFER_SERVERLESS_API_KEY` is unset) |
+
+## Configuration
+
+Add to your pi configuration for automatic loading:
+
+```json
+{
+  "extensions": [
+    "/path/to/pi-wafer-provider"
+  ]
+}
+```
+
+## License
+
+MIT

package/custom-models.json

@@ -0,0 +1 @@
+[]

package/index.ts

@@ -0,0 +1,400 @@
+/**
+ * Wafer Provider Extension
+ *
+ * Registers Wafer Serverless as a custom provider using the
+ * OpenAI completions API.
+ * 
+ * Model resolution strategy: Stale-While-Revalidate
+ *   1. Serve stale immediately: disk cache → embedded models.json (zero-latency)
+ *   2. Revalidate in background: live API /models → merge with embedded → cache → hot-swap
+ *   3. patch.json + custom-models.json applied on top of whichever source won
+ *   
+ * Merge order: [live|cache|embedded] → apply patch.json → merge custom-models.json
+ *
+ * Provider:
+ *   - wafer-serverless  (WAFER_SERVERLESS_API_KEY, falls back to WAFER_API_KEY)
+ *
+ * Usage:
+ *   # Option 1: Store in auth.json (recommended)
+ *   # Add to ~/.pi/agent/auth.json:
+ *   #   "wafer-serverless":  { "type": "api_key", "key": "your-wafer-serverless-key" }
+ *
+ *   # Option 2: Set as environment variables
+ *   export WAFER_SERVERLESS_API_KEY=your-wafer-serverless-key
+ *   # Falls back to WAFER_API_KEY for backwards compatibility
+ *
+ *   # Run pi with the extension
+ *   pi -e /path/to/pi-wafer-provider
+ *
+ * Then use /model to select available models:
+ *   - Qwen3.5-397B-A17B (262K context)
+ *   - GLM-5.1 (202K context)
+ */
+
+import { getAgentDir, type ExtensionAPI, type ModelRegistry } from "@earendil-works/pi-coding-agent";
+import modelsData from "./models.json" with { type: "json" };
+import customModelsData from "./custom-models.json" with { type: "json" };
+import patchData from "./patch.json" with { type: "json" };
+import fs from "fs";
+import path from "path";
+
+// ─── Types ────────────────────────────────────────────────────────────────────
+
+interface JsonModel {
+  id: string;
+  name: string;
+  reasoning: boolean;
+  input: string[];
+  cost: {
+    input: number;
+    output: number;
+    cacheRead: number;
+    cacheWrite: number;
+  };
+  contextWindow: number;
+  maxTokens: number;
+  thinkingLevelMap?: Record<string, string | null>;
+  providers?: string[];
+  headers?: Record<string, string>;
+  compat?: {
+    supportsDeveloperRole?: boolean;
+    supportsStore?: boolean;
+    maxTokensField?: "max_completion_tokens" | "max_tokens";
+    thinkingFormat?: "openai" | "zai" | "qwen" | "qwen-chat-template";
+    supportsReasoningEffort?: boolean;
+    supportsZdr?: boolean;
+  };
+}
+
+interface PatchEntry {
+  name?: string;
+  reasoning?: boolean;
+  input?: string[];
+  cost?: {
+    input?: number;
+    output?: number;
+    cacheRead?: number;
+    cacheWrite?: number;
+  };
+  contextWindow?: number;
+  maxTokens?: number;
+  thinkingLevelMap?: Record<string, string | null>;
+  providers?: string[];
+  compat?: Record<string, unknown>;
+}
+
+type PatchData = Record<string, PatchEntry>;
+
+// ─── Patch Application ─────────────────────────────────────────────────────────
+
+function applyPatch(model: JsonModel, patch: PatchEntry): JsonModel {
+  const result = { ...model };
+
+  if (patch.name !== undefined) result.name = patch.name;
+  if (patch.reasoning !== undefined) result.reasoning = patch.reasoning;
+  if (patch.input !== undefined) result.input = patch.input;
+  if (patch.contextWindow !== undefined) result.contextWindow = patch.contextWindow;
+  if (patch.maxTokens !== undefined) result.maxTokens = patch.maxTokens;
+  if (patch.providers !== undefined) result.providers = patch.providers;
+
+  if (patch.cost) {
+    result.cost = {
+      input: patch.cost.input ?? result.cost.input,
+      output: patch.cost.output ?? result.cost.output,
+      cacheRead: patch.cost.cacheRead ?? result.cost.cacheRead,
+      cacheWrite: patch.cost.cacheWrite ?? result.cost.cacheWrite,
+    };
+  }
+  if (patch.compat) {
+    result.compat = { ...(result.compat || {}), ...patch.compat };
+  }
+
+  if (!result.reasoning && result.compat?.thinkingFormat) {
+    delete result.compat.thinkingFormat;
+  }
+  if (result.compat && Object.keys(result.compat).length === 0) {
+    delete result.compat;
+  }
+
+  return result;
+}
+
+/** Full pipeline: base models → patch → custom → result */
+function buildModels(base: JsonModel[], custom: JsonModel[], patch: PatchData): JsonModel[] {
+  const modelMap = new Map<string, JsonModel>();
+
+  for (const model of base) {
+    modelMap.set(model.id, model);
+  }
+
+  for (const [id, patchEntry] of Object.entries(patch)) {
+    const existing = modelMap.get(id);
+    if (existing) {
+      modelMap.set(id, applyPatch(existing, patchEntry));
+    }
+  }
+
+  for (const model of custom) {
+    const existing = modelMap.get(model.id);
+    const patchEntry = patch[model.id];
+    if (existing && patchEntry) {
+      modelMap.set(model.id, applyPatch(model, patchEntry));
+    } else if (existing) {
+      modelMap.set(model.id, model);
+    } else if (patchEntry) {
+      modelMap.set(model.id, applyPatch(model, patchEntry));
+    } else {
+      modelMap.set(model.id, model);
+    }
+  }
+
+  return Array.from(modelMap.values());
+}
+
+/** Filter models to only those belonging to the given provider, then strip the `providers` field. */
+function filterModelsForProvider(models: JsonModel[], providerId: string): JsonModel[] {
+  return models
+    .filter((m) => !m.providers || m.providers.includes(providerId))
+    .map(({ providers, ...rest }) => rest);
+}
+
+/** Apply per-model ZDR header unless the model explicitly opts out. */
+function applyZdrHeaders(models: JsonModel[]): JsonModel[] {
+  return models.map((model) => {
+    // default: ZDR supported; omit header only if compat.supportsZdr === false
+    if (model.compat?.supportsZdr === false) return model;
+    return {
+      ...model,
+      headers: { ...(model.headers || {}), "Wafer-ZDR": "required" },
+    };
+  });
+}
+
+// ─── Stale-While-Revalidate Model Sync ────────────────────────────────────────
+
+const BASE_URL = "https://pass.wafer.ai/v1";
+const MODELS_URL = `${BASE_URL}/models`;
+const CACHE_DIR = path.join(getAgentDir(), "cache");
+const LIVE_FETCH_TIMEOUT_MS = 8000;
+
+interface ProviderConfig {
+  providerId: string;
+  apiKeyEnv: string;
+  fallbackApiKeyEnv?: string;
+}
+
+/** Transform a model from the Wafer /v1/models API. */
+function transformApiModel(apiModel: any): JsonModel | null {
+  return {
+    id: apiModel.id,
+    name: apiModel.id,
+    reasoning: false,
+    input: ["text"],
+    cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+    contextWindow: apiModel.max_model_len || 0,
+    maxTokens: 0,
+    compat: {
+      supportsZdr: apiModel.zdr_supported ?? undefined,
+      supportsReasoningEffort: true,
+    },
+  };
+}
+
+function getCachePath(providerId: string): string {
+  return path.join(CACHE_DIR, `${providerId}-models.json`);
+}
+
+async function fetchLiveModels(apiKey: string, signal?: AbortSignal): Promise<JsonModel[] | null> {
+  try {
+    const response = await fetch(MODELS_URL, {
+      headers: { Authorization: `Bearer ${apiKey}` },
+      signal: signal ? AbortSignal.any([AbortSignal.timeout(LIVE_FETCH_TIMEOUT_MS), signal]) : AbortSignal.timeout(LIVE_FETCH_TIMEOUT_MS),
+    });
+    if (!response.ok) return null;
+    const data = await response.json();
+    const apiModels = Array.isArray(data) ? data : (data.data || []);
+    if (!Array.isArray(apiModels) || apiModels.length === 0) return null;
+    return apiModels.map(transformApiModel).filter((m): m is JsonModel => m !== null);
+  } catch {
+    return null;
+  }
+}
+
+function loadCachedModels(providerId: string): JsonModel[] | null {
+  try {
+    const data = JSON.parse(fs.readFileSync(getCachePath(providerId), "utf8"));
+    return Array.isArray(data) ? data : null;
+  } catch {
+    return null;
+  }
+}
+
+function cacheModels(providerId: string, models: JsonModel[]): void {
+  try {
+    fs.mkdirSync(CACHE_DIR, { recursive: true });
+    fs.writeFileSync(getCachePath(providerId), JSON.stringify(models, null, 2) + "\n");
+  } catch {
+    // Cache write failure is non-fatal
+  }
+}
+
+function mergeWithEmbedded(liveModels: JsonModel[], embeddedModels: JsonModel[]): JsonModel[] {
+  const embeddedMap = new Map(embeddedModels.map(m => [m.id, m]));
+  const seen = new Set<string>();
+  const result: JsonModel[] = [];
+  for (const liveModel of liveModels) {
+    const embedded = embeddedMap.get(liveModel.id);
+    seen.add(liveModel.id);
+    if (embedded) {
+      // Self-heal: live API pricing is authoritative field-by-field. Prefer the
+      // live cost when the API reports it (non-zero); fall back to embedded when
+      // the API is silent (0) so curated cacheRead/cacheWrite isn't clobbered and
+      // providers whose /models endpoint exposes no pricing keep their curated
+      // cost. Curation (reasoning/input/compat/name) still wins via ...embedded.
+      result.push({
+        ...liveModel,
+        ...embedded,
+        cost: {
+          input: liveModel.cost.input || embedded.cost.input,
+          output: liveModel.cost.output || embedded.cost.output,
+          cacheRead: liveModel.cost.cacheRead || embedded.cost.cacheRead,
+          cacheWrite: liveModel.cost.cacheWrite || embedded.cost.cacheWrite,
+        },
+        contextWindow: liveModel.contextWindow || embedded.contextWindow,
+      });
+    } else {
+      result.push(liveModel);
+    }
+  }
+  // Append any embedded models that the live API didn't return
+  for (const em of embeddedModels) {
+    if (!seen.has(em.id)) {
+      result.push(em);
+    }
+  }
+  return result;
+}
+
+function loadStaleModels(providerId: string, embeddedModels: JsonModel[]): JsonModel[] {
+  const cached = loadCachedModels(providerId);
+  if (!cached || cached.length === 0) return embeddedModels;
+
+  // Merge embedded models that are missing from cache (newly added models)
+  const cachedMap = new Map(cached.map(m => [m.id, m]));
+  for (const em of embeddedModels) {
+    if (!cachedMap.has(em.id)) {
+      cached.push(em);
+    }
+  }
+  return cached;
+}
+
+async function revalidateModels(providerId: string, apiKey: string | undefined, embeddedModels: JsonModel[], signal?: AbortSignal): Promise<JsonModel[] | null> {
+  if (!apiKey) return null;
+  const liveModels = await fetchLiveModels(apiKey, signal);
+  if (!liveModels || liveModels.length === 0) return null;
+  const merged = mergeWithEmbedded(liveModels, embeddedModels);
+  cacheModels(providerId, merged);
+  return merged;
+}
+
+// ─── Per-Provider State ───────────────────────────────────────────────────────
+
+interface ProviderState {
+  cachedApiKey: string | undefined;
+  revalidateAbort: AbortController | null;
+}
+
+function createProviderState(): ProviderState {
+  return { cachedApiKey: undefined, revalidateAbort: null };
+}
+
+/** Resolve API key from pi auth registry, falling back to env vars. */
+async function resolveApiKey(state: ProviderState, providerId: string, modelRegistry: ModelRegistry, config: ProviderConfig): Promise<void> {
+  state.cachedApiKey =
+    (await modelRegistry.getApiKeyForProvider(providerId))
+    ?? resolveEnvKeyValue(config)
+    ?? undefined;
+}
+
+/** Resolve the actual API key value from environment (for fetch calls). */
+function resolveEnvKeyValue(config: ProviderConfig): string | undefined {
+  const { apiKeyEnv, fallbackApiKeyEnv } = config;
+  return process.env[apiKeyEnv] || (fallbackApiKeyEnv ? process.env[fallbackApiKeyEnv] : undefined);
+}
+
+/** Resolve the $-prefixed env var reference for registerProvider.
+ *  Uses the primary key if set, otherwise the fallback key if set,
+ *  otherwise defaults to the primary key name (pi marks it unresolved). */
+function resolveEnvKeyRef(config: ProviderConfig): string {
+  const { apiKeyEnv, fallbackApiKeyEnv } = config;
+  if (process.env[apiKeyEnv]) return `$${apiKeyEnv}`;
+  if (fallbackApiKeyEnv && process.env[fallbackApiKeyEnv]) return `$${fallbackApiKeyEnv}`;
+  return `$${apiKeyEnv}`;
+}
+
+function registerWaferProvider(
+  pi: ExtensionAPI,
+  config: ProviderConfig,
+  embeddedModels: JsonModel[],
+  customModels: JsonModel[],
+  patches: PatchData,
+): void {
+  const { providerId } = config;
+  const state = createProviderState();
+
+  const staleBase = loadStaleModels(providerId, embeddedModels);
+  const staleModels = filterModelsForProvider(
+    buildModels(staleBase, customModels, patches),
+    providerId,
+  );
+
+  pi.registerProvider(providerId, {
+    baseUrl: BASE_URL,
+    apiKey: resolveEnvKeyRef(config),
+    api: "openai-completions",
+    models: applyZdrHeaders(staleModels),
+  });
+
+  pi.on("session_start", async (_event, ctx) => {
+    state.revalidateAbort?.abort();
+    state.revalidateAbort = new AbortController();
+    const signal = state.revalidateAbort.signal;
+    resolveApiKey(state, providerId, ctx.modelRegistry, config).then(() => {
+      revalidateModels(providerId, state.cachedApiKey, embeddedModels, signal).then((freshBase) => {
+        if (freshBase && !signal.aborted) {
+          pi.registerProvider(providerId, {
+            baseUrl: BASE_URL,
+            apiKey: resolveEnvKeyRef(config),
+            api: "openai-completions",
+            models: applyZdrHeaders(
+              filterModelsForProvider(buildModels(freshBase, customModels, patches), providerId),
+            ),
+          });
+        }
+      });
+    });
+  });
+
+  pi.on("session_shutdown", () => {
+    state.revalidateAbort?.abort();
+  });
+}
+
+// ─── Provider Definitions ────────────────────────────────────────────────────
+
+const PROVIDER: ProviderConfig = {
+  providerId: "wafer-serverless",
+  apiKeyEnv: "WAFER_SERVERLESS_API_KEY",
+  fallbackApiKeyEnv: "WAFER_API_KEY",
+};
+
+// ─── Extension Entry Point ────────────────────────────────────────────────────
+
+export default function (pi: ExtensionAPI) {
+  const embeddedModels = modelsData as JsonModel[];
+  const customModels = customModelsData as JsonModel[];
+  const patches = patchData as PatchData;
+
+  registerWaferProvider(pi, PROVIDER, embeddedModels, customModels, patches);
+}