v1.6.0

✨ Added

• Compare over time — a Trend page charts one model's speed across its history. The verdict is judged against that model's own run-to-run spread, so a 5% dip on a steady model is reported and a 23% dip on a jumpy one is not: it says "regression" only when the change exceeds the noise it already shows, with a 5% floor so three near-identical runs can't make anything look significant
• Inter-token latency — the gap between consecutive tokens is recorded during streaming and reported as a median plus a 95th percentile. TPS averages the whole run, so a model that starts fast and then crawls still looks healthy; the p95 is what exposes it, and a column whose p95 is more than 2× its median now shows a marker. Stored in the history DB (schema v5), so it survives a reload and a backup round-trip
• 128K context sweep — the sweep now offers 128K alongside 2K…64K for long-context models. It stays out of the default set: it is slow, and most local models cannot hold it
• KoboldCpp and Jan detection — both are found by the automatic scan (KoboldCpp on 5001 under its /api/v1 OpenAI layer, Jan on 1337). Probed after Ollama, LM Studio and llama.cpp, so an unreachable extra never delays the servers most people run
• Sound effects and desktop notifications — short synthesised cues mark a run starting and finishing (a suite gets its own three-note resolve, an all-error run gets a falling tone), and a desktop notification fires when a long run ends while the window is in the background. Cues are generated with the Web Audio API rather than shipped as audio files, so there are no assets to load and nothing breaks if a webview refuses to start an AudioContext. New "Sound & alerts" section in Settings, with a volume slider that previews the real cue
• Race view — a live timing board above the columns: one lane per model, ranked by tok/s as the run streams, with TTFT split times and the model's own color. Bars encode speed, not characters written, so a model that rambles for 800 tokens doesn't outrank one that answers in 80. Toggled with the Race button in the results toolbar
• Share card (PNG) — saves a run as a fixed-size card (prompt, per-model speed bars, TTFT, token counts, cloud cost) for posting. Renders on a fixed dark ground so the image looks the same regardless of the sender's theme
• One color per model, everywhere — the columns, results chart, Test Suite scoreboard and history chips now derive a model's color from its name instead of its position, so adding a column or shuffling for blind mode no longer repaints the comparison. Colors are assigned collision-free within a run
• Per-model generation parameters — any Bench column can override temperature, top-p and max output tokens from the column header (a dot marks overridden columns; unset fields still inherit the global Advanced settings), so one model can be compared against itself at different temperatures
• Re-run from History — a run can be started again straight from its history card, restoring the prompt, columns, global sampling settings and per-model overrides that produced it
• JSON Schema checks for Test Suite outputs (alongside exact/contains/regex/JSON), with path-level errors and a compact validator covering type, required, properties, items, enum, const, bounds and allOf/anyOf/oneOf/not
• Context sweep — runs one model against the same prompt at 2K…64K context windows (neutral filler, fixed 256-token answers) and charts tok/s and TTFT against window size, with the percentage drop versus the smallest window
• Cloud spend dashboard — new Spend page with total cost, average per run, a 30-day cost trend, and per-provider / per-model breakdowns built from the cost frozen into each run (local runs and runs without exact tokens stay out)
• Bundled fonts — Inter (UI), JetBrains Mono (metrics) and Vazirmatn (Persian) now ship with the app instead of relying on whatever the OS has installed, so typography and Persian rendering are identical on every machine
• Per-connection URL variables — some providers can't be reached from a static base URL. Cloudflare Workers AI needs the account id in the path (…/accounts/{ACCOUNT_ID}/ai/v1), so the Add/Edit key dialog now renders an Account ID field (with the dash.cloudflare.com hint) under the key, and the value is stored per connection. Every request — test, model refresh, per-model probe, bench and suite runs — interpolates it, and a connection without one is skipped rather than sent to a URL that can only 404
• 18 more cloud providers — Cloudflare Workers AI, Venice AI, SiliconFlow, Hugging Face Inference Providers, Hyperbolic, Featherless, Nous Research, LLM7, Morph, Cerebras, Fireworks AI, Alibaba DashScope, Baidu Qianfan, Tencent Hunyuan, MiniMax, Dahl Inference and Cohere, pushing the catalog from 19 to 36 entries. All ride the existing Bearer-auth streaming client; icons drop into public/providers/<id>.png
• Low-sample badge and Wilson bound on the leaderboard — a model with fewer than 5 successful runs is flagged instead of letting a 1/1 record top the board, and each win rate now shows its Wilson 95% lower bound next to the raw percentage

🐛 Fixed

• URL-variable values could escape their segment: the Cloudflare account-id field is substituted into the base URL, and any value containing /, ? or # would have retargeted the request path or injected a query. Values are now validated against the RFC 3986 unreserved set (-._~A-Za-z0-9) and percent-encoded; anything else leaves the placeholder, blocks Save/Check, and is reported in the field instead of producing a silent 404
• Backup restore was lossy: importing a history backup dropped each result's reasoning trail, and would have dropped inter-token latency too. A round-trip now preserves every field the app can store
• Phantom servers in the detection scan: a service answering 200 with unrelated JSON was reported as a local server with an empty model list. With five ports now probed, that was much likelier to happen; a candidate must return at least one model to count as detected
• AMOLED flattened every translucent panel: the theme matched compiled utility classes with [class*='dark:bg-slate-900'] { … !important }, which also caught bg-slate-900/60 and forced it solid — 38 translucent surfaces (cards, drawers, headers) lost their opacity in AMOLED. The theme now remaps Tailwind's own --color-slate-* variables, so color-mix() keeps working, no !important is involved, and new components theme correctly without knowing AMOLED exists. Chrome colour is unchanged: pure black canvas, same surface steps
• Overscroll "slack" at the window edges: dragging with a touchpad or touchscreen bounced/slid the whole page and revealed a gap behind the app. WebView2 swipe-navigation and elastic overscroll are now disabled, the document can no longer scroll, and inner panes contain their own overscroll
• Settings panel never animated: it used animate-in / slide-in-from-right, which come from tw-animate-css — a package this project doesn't install, so the classes silently did nothing. Replaced with a real keyframe that also mirrors under RTL
• Delete button inside a button in Quick Prompts: user presets wrapped a delete button in the preset button, which is invalid HTML and left the delete action unreachable by keyboard. The pill is now a container with two sibling buttons
• Untranslated updater dialog and model picker: updater.* and bench.selectModel were referenced in code but existed in no locale, so those strings fell back to English in all six languages

🔧 Changed

• Prettier is actually enforced now — .prettierrc existed since 1.0 but was never applied, so prettier --check . failed on 40 of ~120 files. The whole tree is formatted once, npm run format / format:check exist, and CI gained a Prettier gate next to ESLint
• 1.6.0 has a date and a clean section — [Unreleased] became [1.6.0] - 2026-10-24, the K127 typo became 128K, the provider count (19 → 36) and the missing Share card / Race / Trend / ITL / notifications entries were added, and README/README.ru/README.zh/SECURITY.md now agree with the code
• Race view re-renders 40% less while streaming — the board's clock ticked every 120 ms and re-derived every lane on each tick, and a fresh modelColorHex() call per lane per render. Ticks now land on a coarser clock, rows are memoized (LaneRow) so only lanes whose numbers moved repaint, and colors resolve once per lane identity
• Font licences ship with the build — Inter, JetBrains Mono and Vazirmatn are OFL 1.1, which requires the licence to travel with the font. The fonts were copied into dist/assets/ but their LICENSE files were not, so an installer shipped fonts without the notice. scripts/copy-font-licenses.cjs (part of npm run build) now writes them to dist/fonts-licenses/, and THIRD_PARTY_NOTICES.md indexes them
• Notification permission is requested at opt-in, not mid-run — turning on desktop notifications in Settings now asks the OS for permission right then; a run finishing in the background only posts if it was already granted. Previously the request fired from the run-completion callback, so the one-shot system dialog appeared at the worst possible moment and a denial burned the only chance the OS gives
• Settings has a manual "Check for updates" — the updater only ran once per launch, so verifying a release meant restarting and hoping. The new button reports all four outcomes distinctly (update available / already latest / check failed / desktop-app only), which is what makes "yes, the endpoint is reachable and signed" a testable result
• Code highlighting no longer ships 450 grammars — PrismAsyncLight imports refractor/all, so the bundler emitted one chunk per Prism grammar: 1057 JS f...

v1.5.0

Polish & reliability release: clearer first-run and streaming feedback, shareable
test suites, sharper benchmark tables, and a roomier history panel.

✨ Added

• Auto-detect local servers — scans Ollama (11434), LM Studio (1234), and llama.cpp (8080) with one-click connect chips in Settings
• First-run onboarding card (Bench) when no local or cloud model is available: live scan results plus one-click install links for Ollama / LM Studio
• Shareable test suites — export a suite (prompts + auto-checks + selected models) to JSON and import it back (suiteFile format, versioned) so benchmarks are reproducible across machines
• Live streaming feedback in each column: a phase strip (Preparing → Generating) with a running elapsed timer, a live ~tok/s estimate, and a typing caret, so a slow first token never looks like a hang
• Bench verdict bar — fastest / lowest-TTFT / cheapest model at a glance, plus the slowest model's deficit as a percentage
• Unified toast notifications for copy, import, and export feedback (replaces the per-component "Copied!" states)
• Test Suite scoreboard upgrades — heat-coloring by relative per-test speed, explicit tok/s units, and a 👑 fastest-per-test marker
• Highlighted code blocks — language label plus a one-click copy button on every fenced block

🔧 Changed

• History sidebar redesigned: wider panel, roomier cards with wrapping model chips (no truncation), and Copy JSON / Export CSV / full-database backup & restore moved into the panel header
• Result toolbar now holds per-run actions only (Copy Report, PDF, Diff); JSON/CSV export lives in the history panel
• RTL-correct logical positions in the leaderboard and history panels (start/end with border-s/border-e)
• Durations under a second now render in milliseconds; TTFT and total duration share one formatter
• Tabular (fixed-width) numerals across metrics, leaderboard, history, and scoreboard so digits no longer jitter
• Syntax highlighting is code-split and only runs after a stream finishes (and for blocks under 20k chars), so it never competes with local inference

🐛 Fixed

• Cloud results were briefly overwritten by a trailing animation frame — pending requestAnimationFrame writes are now cancelled on completion, so a finished cell can no longer flip back to "running"
• keyring 4 renamed delete_password → delete_credential (build fix)

📦 Dependencies

• Added react-syntax-highlighter (lazy-loaded) for code highlighting
• lucide-react 1.39 → 1.56, i18next 26.4.3 → 26.4.4, postcss 8.5.26 → 8.5.30, TypeScript 5.8.3 → 5.9.3, keyring 2 → 4, serde_json 1.0.152 → 1.0.154

v1.4.0

Quality-evaluation release: blind voting that actually works, automatic suite
checks, an LLM judge, OS-keychain secrets, and Elo rankings.

✨ Added

• Blind voting mode: toggle in the footer (before running), auto-shuffle of columns on every blind run, identities masked everywhere, auto-reveal on vote
• Test Suite automatic checks (contains/regex/exact/valid-JSON) with per-cell verdicts, detail badges, and pass-rate summary
• LLM-as-judge panel: blind scoring against a fixed rubric (accuracy/completeness/clarity), deterministic settings, defensive JSON parsing, advisory pick badge that never auto-votes
• OS-keychain storage for API keys (keyring crate + 3 Tauri commands): auto-migration, in-memory cache, silent fallback; secrets centralized in services/secrets.ts
• Leaderboard Elo ratings from pairwise vote outcomes (a 4-model win moves more points than head-to-head)
• Thinking traces persist (reasoning DB column, schema v4) and appear in suite detail, Markdown/PDF exports, and history reloads

🔧 Changed

• ESLint (type-aware + react-hooks) + Prettier configs, zero warnings/errors enforced in CI
• Release workflow: code-signing scaffolding (commented, awaiting certificates)
• History cards use explicit source/provider_id columns (local openai/* models no longer misread as cloud)
• TypeScript major updates ignored by Dependabot (incompatible with typescript-eslint 8)
• jsPDF/html2canvas load lazily on export (main chunk 1.49MB → 1.06MB)
• Release notes drafted from CHANGELOG via scripts/extract-notes.cjs

🐛 Fixed

• Keychain service namespace fixed in Rust; secret-storage badge (OS keychain vs local) in Cloud header

v1.3.0

[1.3.0] - 2026-10-13

✨ Added

• N-repeat runs (1–5, medians reported) with mean±stdev tooltips; exact Ollama load/prompt timings; defensive llama.cpp/LM Studio rate parsing
• Thinking-trace display for reasoning models (reasoning_content collapsible)
• Sequential/Parallel run mode (sequential default: uncontaminated local TPS)
• Server-defaults toggle + seed/max-tokens controls (Ollama num_predict mapped)
• Script-aware token heuristic (CJK ≈ 1.5, Arabic-script ≈ 2.5 chars/token)
• Test Suite "fastest" wording; run-config snapshot (run_config column) with card tooltip
• SQL-side history search + pagination with "Show more"; full-DB JSON backup/export-import
• Shared runTarget streaming core; Test Suite survives page switches and saves runs to history; bench retry updates its history row
• Leaderboard Wilson ranking over successful runs only; ConfirmModal replaces native dialogs; JSON copy empty-state
• Text-based PDF (selectable, paginated) with image fallback for non-Latin scripts
• Signed in-app updates (tauri-plugin-updater): startup check against GitHub releases with consent dialog, progress, and relaunch; release workflow signs artifacts
• Secrets centralized in services/secrets.ts (documented OS-keychain swap plan)
• Community features by contributors: relative timestamps on history cards (@AyanJaved), winners-only filter, reset-to-defaults (@lb1192176991-lab), Ctrl+K prompt focus (@lb1192176991-lab), Test Suite retry-failed (@afifm561)

🔧 Changed

• Explicit cloud: model prefix + source/provider_id history columns (no more local/cloud name collisions)
• Versioned DB migrations (PRAGMA user_version); transactional saves; single-JOIN history; run_id index
• Markdown img is click-to-load and links open externally; restrictive Tauri CSP
• CSV: BOM, formula-injection guard, full (unlimited) export
• CI runs unit tests (npm test, 49 tests); clippy -D warnings enforced, cargo fmt no longer best-effort; npm audit clean (0 vulnerabilities)
• .gitattributes/.editorconfig, Dependabot, SECURITY.md versions, Cargo metadata
• Missing de history keys + code-default-only cloud keys added to all locales
• Remaining hardcoded UI strings moved to i18n (stop marker, add-model button, PDF error, print report, provider form)

🐛 Fixed

• Stream finalize without [DONE] (llmService + ollamaService); mid-stream Ollama errors surfaced; JSON error bodies on HTTP failures
• loadModels preserves cloud slots; Ctrl+Enter gated to Bench; saveHistory returns lastInsertId (vote race removed)

v1.2.0

[1.2.0] - 2026-09-27

Nine new cloud providers (19 total), three new locales, GPL-3.0, and community
files — plus robustness fixes for providers with non-standard APIs.

✨ Added

• 9 new cloud providers: Kilo Gateway, Bazaarlink, BytePlus ModelArk, Poolside, Agnes, Xiaomi MiMo, Mistral, B.AI, and Anthropic (via its official OpenAI SDK compatibility layer, Bearer auth) — each with brand icon, info banner, suggested models, and a direct API-key link
• Real API-key links everywhere: the "Where to get an API key" button now points at each provider's actual key-creation page and opens it in the system browser via the Tauri opener plugin (new openExternal helper; plugin registered in lib.rs)
• Support for providers without a /models endpoint (e.g. Poolside): key test falls back to a key-only probe, reports modelsListed: false, and the detail page shows a hint to add model ids manually instead of an error
• Actionable network errors: DNS/reset/region-blocked failures now explain themselves ("check connection/VPN — some providers block certain regions") instead of a bare Failed to fetch
• 3 new locales: French, Simplified Chinese, and Russian — 6 languages total (EN, FA, DE, FR, ZH, RU) with a 6-tile language picker
• Community files: CODE_OF_CONDUCT.md (Contributor Covenant 2.1), SECURITY.md (supported versions, private reporting, scope, key-storage limitation), CONTRIBUTING links to both
• Credits: "Powered by Sina Cadman" footer in the sidebar and settings panel, linking to GitHub
• Backdrop dim for History and Leaderboard panels (same as Settings), click-to-close

🔄 Changed

• License: MIT → GPL-3.0-only (LICENSE, package.json, README)
• Settings provider form restacked vertically for the panel layout

🐛 Fixed

• Local model picker stuck on the first model: decodeValue returned the raw JSON option value instead of the parsed model name for local sources (cloud sources were unaffected)
• Provider icon fallback tile rendered white-on-white (invisible initials) after a styling change — brand color restored, white backdrop kept only behind raster icons

v1.1.0

[1.1.0] - 2026-09-19

Cloud release: manage cloud LLM providers 9Router-style and benchmark their models side-by-side with local ones — in both Bench and Test Suite.

✨ Added

• Cloud Providers page (new sidebar entry): Custom Providers section (user-added OpenAI-compatible endpoints) plus a curated catalog split into Free Tier Providers and API Key Providers, with search, per-section Test All, and per-card connection status
• 10 providers in the catalog: OpenRouter, DeepSeek, Groq, Together AI, OpenAI, xAI (Grok), Kimi (Moonshot), NVIDIA NIM, Ollama Cloud, and Gemini (via Google's OpenAI-compatibility endpoint) — each with docs link, info banner, suggested models, and a 128px brand icon (public/providers/, graceful initials-tile fallback)
• Provider detail page: named API-key connections (add / test / edit / delete / enable-toggle), an Available Models section that fetches the real /models list, model search, per-model enable toggles, per-model 1-token probe test (green check / red X with latency or error), copy-id, remove, manual add, suggested-model chips, and enable/disable-all
• Cloud models in Bench: grouped model picker (Local + ☁ per-provider groups) with a cloud badge on cloud columns; runs use the stored key via Authorization: Bearer header. Cloud models are stored as qualified names (providerId/model), so history, leaderboard, votes, charts, diff, and all exports work with no database migration
• Cloud models in Test Suite: per-provider cloud checkboxes (mixable with local models in one suite); suites run without any local model installed
• LLMService.testCloudConnection now also returns the model id list; new probeCloudModel for single-model reachability checks

🔄 Changed

• Cloud store migrated to v2 (multiple named keys per provider, enabledModels, knownModels) with automatic non-destructive migration of v1 connections
• Add-model button no longer depends on local models — with cloud-only sources it adds the first enabled cloud model instead of staying disabled
• Removed obsolete nav.soon / *Soon i18n keys (Cloud and Tests are real pages now); full EN + FA + DE coverage for all new UI

🐛 Fixed

• Missing X icon import in ProviderDetailPage that broke tsc (probe-error state)
• Test Suite Run stayed disabled when only cloud models were selected (leftover local-models guard)

v1.0.0

[1.0.0] - 2026-09-17

First stable release. The headline changes are a full App Shell redesign that frees up space for new workflows, a brand-new Test Suite page for batch benchmarking, an AMOLED Dark theme, and a proper language picker — plus German locale coverage and CI updates.

✨ Added

• Test Suite (batch execution): run a list of 1–10 test prompts across all selected models sequentially, with a live progress bar and Stop support. Results land in a scoreboard matrix (one row per test, one column per model) showing TPS per cell, a 👑 crown for the fastest model on each test, an average-TPS + wins summary row, click-to-read full-output detail modal, Markdown report copy, and CSV download. The suite definition persists in localStorage
• AMOLED Dark theme: pure-black (#000000) canvas with neon glows on primary actions, status dots, and focus rings. Builds on top of dark mode, so every dark: style keeps working; persisted in settings (store migrated to v2, old dark/light values carry over untouched)
• Language picker modal: grid-style popup (region code + native name tiles, e.g. US/English, DE/Deutsch, IR/فارسی) with selected-state highlight, Esc/backdrop close, and support for future locales — opened from both the header and the settings panel
• Appearance section in settings: explicit Light / Dark / AMOLED picker alongside the header's three-state theme cycle button
• German (Deutsch) locale: full UI coverage in English, Persian (Farsi, RTL), and German
• App Shell navigation: icon-rail sidebar (Bench, Tests, History, Board, Config) with a reserved "Cloud — Soon" slot for the upcoming cloud-input feature; History, Leaderboard, and Settings now open as sidebar-driven overlay panels instead of crowding the header

🔄 Changed

• Header decluttered: the old 10+-button wrapping header is now a slim 48px TopBar (title + version, provider connection pill with status dot and model count, language, theme, refresh, settings). All result actions (Copy Report, PDF, Diff, Copy JSON, Export CSV) moved to a contextual ResultToolbar that only appears when there are results
• Settings live in a slide-over panel: provider/connection, appearance, language, and advanced model controls (system prompt, temperature, top-p, context length) moved out of the stacked inline layout, giving model outputs the maximum vertical space
• Prompt dock restyled: floating-card prompt box with Ctrl+Enter hint and a dedicated Run/Stop button
• Test Suite owns its page: the sidebar Tests entry is now a real page (bench prompt dock and result toolbar are hidden while it is active)
• CI and release workflows now build with Node.js 24 (was 20)
• Version centralized as before (src/version.ts ↔ package.json ↔ tauri.conf.json ↔ Cargo.toml), now at 1.0.0

🐛 Fixed

• Language modal opened from the header rendered clipped inside the header bar (missing backdrop and title). Root cause: the header's backdrop-blur creates a containing block for fixed descendants — the modal now renders via a React portal to document.body, so it always covers the viewport
• Test Suite worker cell updates could target the wrong cell under concurrent callbacks — cell execution now carries its test index explicitly instead of a shared mutable ref

v0.3.0

[0.3.0] - 2026-09-15

✨ Added

• Side-by-side diff view: compare any two model outputs line-by-line, with similarity percentage, added/removed line counts, pair picker (for 3–4 model runs), and a toggle to hide unchanged lines — no new dependencies (custom LCS-based diff)
• Winner votes: crown the best answer per run with a per-column vote button; the winner gets a highlighted column, a badge, and is remembered in history
• Model leaderboard: aggregated per-model stats (run count, average TPS, win count, win rate) in a sortable sidebar panel, persisted in SQLite
• CSV export: download the full benchmark history as CSV (one row per model result: run id, date, prompt, model, TPS, estimated flag, TTFT, winner, output)
• Keyboard shortcuts: Ctrl+Enter (or Cmd+Enter) to run, Esc to stop generation or close panels

🔄 Changed

• Unified TPS definition: TPS now measures the generation phase only (excluding time-to-first-token) on every provider, matching the meaning of Ollama's eval_count/eval_duration. When a server doesn't report token usage (some LM Studio / llama.cpp setups), TPS is estimated from output length and clearly marked with ~ in columns, charts, history, and Markdown/PDF exports
• Settings now persist: theme, language, provider, base URL, and advanced controls (system prompt, temperature, top-p, context length) survive app restarts via persisted store
• Database schema extended (non-destructive, auto-migrated on launch): new winner_model column on benchmark_runs, plus tps_estimated and token_source on benchmark_results. Databases from v0.2.0 open without data loss
• Version is now centralized in src/version.ts (kept in sync with package.json, tauri.conf.json, and Cargo.toml); app window title and default size updated (1280×800, min 1024×640)

🐛 Fixed

• Run button could get stuck in the "running" state forever when any single model column errored — completion is now tracked per-slot (success, error, or stop), partial runs are always saved to history, and failed/stopped columns get a dedicated Retry button that re-runs only that slot
• Switching providers no longer wipes a manually edited server URL unless the provider actually changed
• JSON.parse on Ollama NDJSON streams is now guarded — a single malformed line no longer kills the whole stream
• Removed unused i18n key (app.version) that showed a stale hardcoded version in the header

v0.2.0

Changelog

All notable changes to this project will be documented in this file.

[0.2.0] - 2026-09-05

✨ Added

• Multi-provider support: connect to LM Studio or any llama.cpp server (OpenAI-compatible API) in addition to Ollama, with a provider/base-URL switcher in the header
• Compare 3–4 models simultaneously (previously limited to 2), with dynamic add/remove model columns
• User-savable prompt presets: save your own prompts alongside the built-in quick presets, with delete support
• TPS / TTFT comparison chart: visual bar charts comparing tokens/sec and time-to-first-token across all running models
• Rich PDF export: generate a fully-formatted PDF report (with Markdown rendering, tables, and code blocks preserved) of any benchmark run
• Full English + Persian (Farsi) i18n coverage across the entire UI, including the history sidebar and export dialogs

🔄 Changed

• Database schema redesigned to support an arbitrary number of models per run (benchmark_runs + benchmark_results tables replace the old fixed model1/model2 history table). Existing history is automatically migrated on first launch.
• History sidebar now respects the active light/dark theme (previously always dark)
• Benchmark chart tooltips are now readable in all theme combinations

🐛 Fixed

• Run/Stop button could get stuck in the "running" state after a completed or failed generation
• Division-by-zero when calculating tokens/sec for near-instant responses
• PDF export now shows clear loading and success feedback, preventing accidental duplicate downloads

---

[0.1.0] - 2026-09-04

• Initial release: side-by-side Ollama model benchmarking, streaming output, TPS/TTFT metrics, local SQLite history, Markdown rendering, dark/light theme, basic prompt presets

v0.1.0 — Initial Release

🎉 First Release — PromptDeck v0.1.0

Local, offline benchmarking studio for Ollama models — compare two models side-by-side in real time.

Features

• Side-by-side streaming comparison of two Ollama models
• Live TPS / TTFT / duration metrics
• Local SQLite history with browsable sidebar
• Markdown rendering with GFM support
• Advanced controls (system prompt, temperature, top-p, context length)
• Dark/light theme, English + Persian (RTL) i18n

Requirements

• Ollama installed and running locally
• At least one model pulled (e.g. ollama pull llama3)

See the README for setup instructions.