ORAG 0.6.0-alpha.12

Draft for v0.6.0-alpha.12: see CHANGELOG.md. Check a download with SHA256SUMS or gh attestation verify <file> --repo orkono/orag (docs/signing.md).

ORAG 0.5.0

[0.5.0] - 2026-10-20
ORAG v0.5.0. The setup page shows this computer and the answer models that fit it, and downloads the one the user picks: the default qwen3.5-4b-q4_k_m (recommended) or qwen3.5-9b-q4_k_m (larger, slower). Each tier has a RAM floor and a recommended RAM measured on Mac (Metal) and Linux (CPU); a hand-set generator is kept; the choice is written to config.toml only after the download verifies (steps 81-82). Packages are still unsigned.
Release gates on the candidate 0.5.0-alpha.1: real-model check unchanged from v0.4.1 (v02 hybrid in context 0.931, hybrid p95 14.4 ms on Metal; answers expected facts 0.925, abstention P/R 0.944/0.850, no uncited answer; seed recall@10 1.000); D-003 dense search at 100k p95 97.3 ms on Apple M5 Max and 226.3 ms on the Linux reference (both PASS). Answer time on a 4-vCPU CPU without a GPU is about 108 s (4B) and 194 s (9B), so GPU support for Windows and Linux is the next step.

ORAG 0.4.1

[0.4.1] - 2026-10-20
ORAG v0.4.1. The web UI and the desktop app's status page speak Turkish and English, with a TR/EN menu; the page opens in Turkish unless English was chosen (step 78). Every release now carries SHA256SUMS and GitHub build provenance for each file, and the macOS desktop build signs and notarizes once the owner adds the Apple secrets (step 79, docs/signing.md); the packages of this release are still unsigned. The desktop app no longer reads its window's address, which could crash it on macOS before a page had loaded.
Release gates on the candidate 0.4.1-alpha.2: real-model check unchanged from v0.4.0 (v02 hybrid in context 0.931, hybrid p95 14.2 ms on Metal; answers expected facts 0.925, abstention P/R 0.944/0.850, no uncited answer; seed recall@10 1.000); D-003 dense search at 100k p95 104.5 ms on Apple M5 Max and 222.4 ms on the Linux reference (both PASS). The release workflow's macOS desktop smoke passed on GitHub.

ORAG 0.4.0

[0.4.0] - 2026-10-19
ORAG v0.4.0. The setup page downloads the models with one button (step 76): pinned Hugging Face revisions, SHA-256 checked, resumable, with progress, cancel and a free-space check; ORAG restarts by itself under orag start and in the desktop app, so a new install needs no hand-made model packs. Offline packs still work, and models.download = false turns the download off. Five code-review rounds (gpt-6-astra) hardened cancel, links in the models folder, proxies and TLS time limits.
Release gates on the candidate 0.4.0-alpha.1: real-model check unchanged from the recorded baselines (v02 hybrid in context 0.931, hybrid p95 14.2 ms on Metal; answers expected facts 0.925, abstention P/R 0.944/0.850, no uncited answer; seed recall@10 1.000); D-003 dense search at 100k p95 96.4 ms on Apple M5 Max and 225.7 ms on the Linux reference (both PASS). The packages are still unsigned. D-020: release tags are now pushed by the agent (owner decision).

ORAG 0.3.0

[0.3.0] - 2026-10-15

• ORAG v0.3.0. A desktop application for macOS, Windows and Linux (Tauri, D-016) runs the service in its own data folder, shows the page in its own window, restarts the service after setup and after crashes, and leaves no service behind when it closes or is killed (docs/desktop.md). The engine gains cross-language fusion (step 48), Turkish stem terms in lexical search (step 51), heading detection in plain text and PDF (step 52) and an uncited flag on answers that cite no source (step 67).
• Evaluation: the 200-question v02 set was rewritten by one model family and audited by another (step 49); scorer defects were fixed (steps 54, 56, 58); orag eval retrieval measures the prompt context exactly as the answer path builds it (step 55); a fresh 130-row v0.4 holdout was added (step 60). The set is still synthetic and not audited by people (D-018).
• Tried and not adopted, each under a rule written before scoring: context diversity (step 50), stricter manual headings (step 53), a 10-chunk context (step 57), bge-m3 as the default embedder (steps 59, 66: better retrieval and faster on CPU, but one unsupported uncited answer), rerankers (steps 61-65: +8 dev questions, but 214 ms or more of latency on Metal and seconds on a CPU), preventing uncited answers at generation time (step 68: no dev target). The default models stay Qwen3-Embedding-0.6B and Qwen3.5-4B.
• Release artifacts: besides the CLI archives and the Windows CLI installer, the draft release carries the desktop app as ORAG_<version>_aarch64.dmg, ORAG_<version>_x64-setup.exe (with the offline WebView2 installer), ORAG_<version>_amd64.AppImage and .deb, each smoke-tested in the release build. Not signed or notarized yet.
• Release gates on the candidate 0.3.0-alpha.28: real-model check unchanged from the recorded baselines (v02 hybrid in context 0.931, hybrid p95 13.9 ms on Metal; answers expected facts 0.925, abstention P/R 0.944/0.850, no uncited answer; seed recall@10 1.000); D-003 dense search at 100k p95 100.8 ms on Apple M5 Max and 208.7 ms on the Linux reference (both PASS).