Skip to content

Releases: Jonathhhan/ofxIC

ofxIC 0.2.0

Choose a tag to compare

@Jonathhhan Jonathhhan released this 27 Aug 15:54

First stable endpoint-first release of ofxIC. The addon connects openFrameworks applications to external inference processes for endpoint inspection, chat, allowlisted local-document search, grounded answers, and explicit media/audio/segmentation protocol adapters without embedding model runtimes. The regular Release GUI path was exercised against a local llama-server with a tool-capable GGUF model; deterministic Windows/Linux tests and the Linux nightly GUI workflow passed from this tag. See docs/RELEASE_NOTES.md and docs/ROADMAP.md for detailed changes, evidence boundaries, and the long-term maintenance plan.

v0.2.0-rc.4

v0.2.0-rc.4 Pre-release
Pre-release

Choose a tag to compare

@Jonathhhan Jonathhhan released this 26 Aug 11:08

Release candidate with operation-scoped timeout and cancellation controls, structured request-failure classification, deterministic lifecycle GUI coverage, reliable multiword chat submission, OpenAI image-size defaults compatible with GPT Image, surfaced provider error details, and a C++20 standalone test baseline. Validation includes Ubuntu/Windows deterministic tests, the openFrameworks nightly build, and deterministic GUI smokes for chat, documents, transcription, SAM, music, native video, timeout, and cancellation. Paid OpenAI and marker-gated Hugging Face live generation were not claimed as release evidence.

ofxIC 0.2.0-rc.3

ofxIC 0.2.0-rc.3 Pre-release
Pre-release

Choose a tag to compare

@Jonathhhan Jonathhhan released this 26 Aug 09:10

Third release candidate for ofxIC 0.2.0.

Added since rc.2

  • Local ACE-Step and hosted Stability Audio 3 music workflows with bounded protocol clients, example controls, timestamped output, playback, and deterministic coverage.
  • A user-invoked, bounded one-URL web snapshot utility with provenance and safe-output tests.
  • A SAM bridge health route, backend inspection, and explicit CPU/CUDA forwarding.

Hardened and fixed

  • Document indexing now enforces bounded documents, sources, chunks, and linear chunk-window lookup while labeling retrieved source text as untrusted evidence.
  • Transcription errors preserve HTTP status and bounded provider details.
  • The regular example groups transcription, image/video, music, and SAM into a visible tabbed Inference tasks workspace with stable preview sizing.

Validation

Marker-gated paid or model-backed live evidence remains separate from deterministic coverage.

See the complete release notes.

ofxIC 0.2.0-rc.2

ofxIC 0.2.0-rc.2 Pre-release
Pre-release

Choose a tag to compare

@Jonathhhan Jonathhhan released this 25 Aug 19:40

Second release candidate for ofxIC 0.2.0.

Fixed since rc.1

  • The regular example now selects https://api.openai.com/v1 immediately when OpenAI audio is active.
  • Settings created before the audio fields existed are migrated to the matching protocol default.
  • Explicit custom and environment-provided transcription endpoints remain authoritative.

Validation

See the complete release notes.

ofxIC 0.2.0-rc.1

ofxIC 0.2.0-rc.1 Pre-release
Pre-release

Choose a tag to compare

@Jonathhhan Jonathhhan released this 25 Aug 12:52

Release candidate for the additive endpoint-first update after v0.1.0.

Highlights

  • Separate chat, transcription, segmentation, and media endpoint configuration.
  • Explicit OpenAI and whisper.cpp transcription clients.
  • Explicit ofxIC SAM bridge v1 with positive and negative multi-point prompts.
  • Hardened request limits, cancellation, credentials, and stable bridge errors.
  • Deterministic GUI coverage for audio, SAM, and video without provider costs.

Validation

Live Whisper and SAM model-backed smokes remain deliberately separate from deterministic fixture evidence.

See the complete release notes.

ofxIC 0.1.0

Choose a tag to compare

@Jonathhhan Jonathhhan released this 24 Aug 15:07
97016a7

First release of ofxIC — Inference Connector for openFrameworks.

ofxIC provides a compact client-side interface for external local and hosted inference endpoints. Model runtimes remain separate processes.

Included

  • OpenAI-compatible endpoint inspection and chat completions
  • Conversation history, streaming, and bounded tool execution
  • Allowlisted local document search with citations
  • OpenAI-compatible image generation
  • Asynchronous "stable-diffusion.cpp" image and video jobs
  • Hugging Face image and queued video inference through fal-ai
  • "ofxImGui" example with selectable endpoint and media profiles
  • Deterministic Linux and Windows tests
  • openFrameworks Nightly build, GUI, image, and video validation

Deliberate boundary

ofxIC does not embed ggml, llama.cpp, CUDA, model files, or other inference runtimes. It connects to them through explicit HTTP endpoints, allowing servers and models to be updated independently.

See the "release notes" (https://github.com/Jonathhhan/ofxIC/blob/v0.1.0/docs/RELEASE_NOTES.md) for the complete scope and deferred features.