@canarycoders/ai 0.5.0 → 0.7.0

package/README.md

@@ -4,7 +4,7 @@ TypeScript SDK for the [CanaryCoders AI](https://ai.canarycoders.es) gateway. Ru
 
 **Docs:** https://docs.ai.canarycoders.es
 
-One client covers chat, image/video/audio generation, embeddings, vision, conversational agents, realtime sessions, and usage data. The queue polling, SSE streaming, retries, and error typing are handled for you, so most calls are a single `await`.
+One client covers chat, image/video/audio generation, embeddings, vision, OCR/PII-redaction, conversational agents, realtime sessions, and usage data. The queue polling, SSE streaming, retries, and error typing are handled for you, so most calls are a single `await`.
 
 ## Install
 
@@ -77,6 +77,20 @@ const song = await client.audio.music({ prompt: "calm lo-fi", durationMs: 20000
 const { detections } = await client.vision.detect({ image });
 ```
 
+## OCR and PII redaction (local)
+
+Text recognition with bounding boxes and PII redaction, processed fully on CanaryLLM's own EU infrastructure (Apple Vision). Images and extracted text are never stored.
+
+```ts
+const { text, lines } = await client.ocr.recognize({ image, granularity: "word" });
+
+const { redactedImage, counts } = await client.ocr.redact({
+  image,
+  presets: ["email", "phone", "iban"],
+  style: "blur", // "box" is the safest for hard PII redaction
+});
+```
+
 ## Embeddings
 
 Embed text into vectors with a local model (LM Studio), for customer-side RAG. The gateway processes the text transiently and stores nothing — you keep the vectors and documents in your own store (e.g. pgvector, sqlite-vec).
@@ -161,6 +175,10 @@ await oa.chat.completions.create({
 | `defaultTag` | — | attached to requests for usage attribution |
 | `poll` | — | default poll timing for queued ops |
 
+## Agent skill
+
+`skills/canary-ai/` ships a Codex/Claude Code Agent Skill that drives the gateway (research, X search, chat, image generation, model discovery) from a single Bun script, no MCP server needed. See [docs.ai.canarycoders.es/agent-skill](https://docs.ai.canarycoders.es/agent-skill) for install/update/uninstall steps. It's a plain repo folder, not part of the published npm package.
+
 ## Keeping types in sync
 
 The gateway API is the source of truth. Regenerate types from its OpenAPI spec and diff them against the hand-written ones:

package/dist/index.cjs

@@ -1588,6 +1588,27 @@ var VisionResource = class extends BaseResource {
   }
 };
 
+// src/resources/ocr.ts
+var OcrResource = class extends BaseResource {
+  recognize(params, poll) {
+    return this.runQueued("/api/ocr/recognize", params, "vision", poll);
+  }
+  recognizeJob(params, signal) {
+    return this.submitQueued("/api/ocr/recognize", params, "vision", signal);
+  }
+  redact(params, poll) {
+    return this.runQueued("/api/ocr/redact", params, "vision", poll);
+  }
+  redactJob(params, signal) {
+    return this.submitQueued("/api/ocr/redact", params, "vision", signal);
+  }
+  async languages(signal) {
+    return this.transport.json("GET", "/api/ocr/languages", {
+      signal
+    });
+  }
+};
+
 // src/client.ts
 var DEFAULT_BASE_URL = "https://api.ai.canarycoders.es";
 function readEnv(key) {
@@ -1602,6 +1623,7 @@ var CanaryCodersAI = class {
   audio;
   embeddings;
   vision;
+  ocr;
   conversations;
   agents;
   realtime;
@@ -1634,6 +1656,7 @@ var CanaryCodersAI = class {
     this.audio = new AudioResource(this.transport, poll);
     this.embeddings = new EmbeddingsResource(this.transport, poll);
     this.vision = new VisionResource(this.transport, poll);
+    this.ocr = new OcrResource(this.transport, poll);
     this.conversations = new ConversationsResource(this.transport, poll);
     this.agents = new AgentsResource(this.transport, poll);
     this.realtime = new RealtimeResource(this.transport, poll);