@fre4x/openai 1.1.0-beta.3 → 1.1.0-beta.4

package/README.md

@@ -4,17 +4,17 @@
 
 Part of **[FRE4X-B1TE](../)** — a monorepo of MCP servers built for autonomous agents.
 
-This B1TE bridges your agent directly to OpenAI's advanced models — enabling it to see with GPT-4o, imagine with DALL-E 3, and hear with Whisper.
+This B1TE bridges your agent directly to OpenAI's advanced models — enabling it to see with GPT-4o, imagine with GPT Image 2, and hear with Whisper.
 
 ## Tools
 
 | Tool | Capability |
 |------|-----------|
-| `openai_analyze_media` | Analyze an image from a `file://` or `http(s)://` URL. Prompt is optional. |
-| `openai_list_models` | List available OpenAI models with pagination. |
-| `openai_generate_image` | Image synthesis via DALL-E 3. Optionally save to `output_dir`. |
-| `openai_transcribe_audio` | Transcribe local audio files using Whisper (`whisper-1`). |
-| `openai_generate_speech` | Generate speech audio from text and optionally save it locally. |
+| `analyze_media` | Analyze an image from a `file://` or `http(s)://` URL. Prompt is optional. |
+| `list_models` | List available OpenAI models with pagination. |
+| `generate_image` | Image synthesis via GPT Image 2. Embeds base64 image data and optionally saves to `output_dir`. |
+| `transcribe_audio` | Transcribe local audio files using Whisper (`whisper-1`) and auto-save a `.txt` transcript. |
+| `generate_speech` | Generate speech audio from text and optionally save it locally. |
 
 ## Setup
 
@@ -42,11 +42,12 @@ This B1TE bridges your agent directly to OpenAI's advanced models — enabling i
 ## Features
 
 - **Protocol Compliance**: Implements the split `content` (Markdown) and `structuredContent` (Raw Data) model for high-fidelity agent interaction.
-- **Agent-Sufficient Output**: `content.text` carries enough detail for autonomous follow-up, including model IDs, generated image URLs, revised prompts, saved paths, transcript metadata, and speech metadata.
+- **Agent-Sufficient Output**: `content.text` carries enough detail for autonomous follow-up, including model IDs, generated image metadata, revised prompts, saved paths, transcript metadata, and speech metadata.
 - **Strict Validation**: All tools utilize `outputSchema` for reliable data parsing and reduced hallucinations.
 - **Privacy & Security**: Aggressively masks implementation details (HTTP codes, library names) in error messages.
 - **Token Efficiency**: Mandatory pagination for model listing.
-- **Vision Input Contract**: `openai_analyze_media` accepts only `file://` and `http(s)://` URLs. `data:` URLs are rejected at validation time.
+- **Vision Input Contract**: `analyze_media` accepts only `file://` and `http(s)://` URLs. `data:` URLs are rejected at validation time.
+- **Transcript Saving**: `transcribe_audio` always writes `openai_transcription_{number}.txt` into the current working directory unless `output_dir` overrides the destination.
 
 ## Development