release: cut v0.1.0 in CHANGELOG

Consolidates every item that was queued under [Unreleased] into a dated
[0.1.0] — 2026-04-24 block organised by theme (core product, CLI &
scripting, website & AI search, repository hygiene). Leaves [Unreleased]
empty for the next cycle. Adds a new footer link for the v0.1.0 compare
URL and retargets the Unreleased compare link from main...HEAD to
v0.1.0...HEAD so future diffs land at the right base.

Author: carrilloapps

CHANGELOG.md

@@ -6,114 +6,129 @@ adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
 
-### Added
+_Nothing yet — open an issue or PR to propose the next thing._
 
-- **`hfo --bench <tag>`**. A standardised 4-prompt benchmark (warmup · code ·
-  reasoning · translation) that runs against a local Ollama model and
-  reports tokens-per-second and time-to-first-token, both per-prompt and
-  aggregate. Uses Ollama's streaming `/api/generate` endpoint and the
+## [0.1.0] — 2026-04-24
+
+First public release of **hfo** (`hfo-cli` on npm).
+
+### Added — core product
+
+- Fullscreen TUI with six tabs (Dashboard, Models, Install, Tune, Help,
+  Settings) on top of a hardware-aware install engine. Alt-screen buffer
+  preserves your scrollback on exit.
+- Hardware scoring: every GGUF in a Hugging Face repo graded 0–100
+  against usable VRAM and RAM, with per-quant labels like `Full GPU`,
+  `Partial 87%`, `CPU-heavy`.
+- Hugging Face model-card parser: pulls recommended `temperature`,
+  `top_p`, `top_k`, `repeat_penalty`, `min_p`, and context size into the
+  generated Modelfile.
+- Quick-confirm install flow: on the happy path a new model installs in
+  three keystrokes — pick a quant, press <kbd>Enter</kbd>, wait. Branches
+  to the file browser (<kbd>O</kbd>) or params editor (<kbd>C</kbd>) only
+  when there's something to decide.
+- Orphan reinstall (<kbd>I</kbd>): tags removed from Ollama but whose
+  GGUFs still live on disk get recovered, regenerating the Modelfile when
+  needed.
+- Zip backup + restore with level-9 compression, streaming (multi-GB
+  safe), sidecar `metadata.json` per backup.
+- Deep delete: <kbd>d</kbd> removes the Ollama tag only, <kbd>Alt</kbd>+<kbd>d</kbd>
+  also wipes the tracked directory.
+- ~90% capacity Ollama env tuner for `OLLAMA_FLASH_ATTENTION`,
+  `OLLAMA_KV_CACHE_TYPE`, `OLLAMA_KEEP_ALIVE`, `OLLAMA_NUM_PARALLEL`,
+  `OLLAMA_MAX_LOADED_MODELS`, `OLLAMA_MAX_QUEUE`. Persists via `setx`
+  (Windows) · `launchctl setenv` + `~/.zprofile` (macOS) · `~/.profile`
+  + systemd override (Linux).
+- 7 themes (Dark, Light, Dracula, Solarized Dark/Light, Nord, Monokai)
+  and 20 language packs with live language switching from Settings.
+
+### Added — CLI & scripting
+
+- Headless flag equivalents for every TUI capability: `--view` · `--list`
+  · `--tune` · `--backup` · `--restore` · `--delete` · `--launch` · `--tab`.
+- Launch matrix `L` / `--launch <integration>` wiring into Claude Code,
+  Cline, Codex, Copilot CLI, Droid, Hermes, Kimi, OpenCode, OpenClaw, Pi,
+  and VS Code, with runtime probing against `ollama launch --help` to
+  mark unsupported targets.
+- **Coding-agent launch manifest + hints.** Every `--launch <agent>`
+  writes a JSON manifest to
+  `<configDir>/hfo/agent-launches/<agent>.json` recording which model was
+  bound to which agent and when, then prints a per-agent hint block
+  listing the env vars / config keys needed to route that agent through
+  the local Ollama (Claude Code, Codex, Cline, OpenCode, Droid, VS Code,
+  Copilot CLI). hfo never mutates user-owned config files — only records
+  and educates.
+- **`hfo --bench <tag>`.** A standardised 4-prompt benchmark (warmup ·
+  code · reasoning · translation) that runs against a local Ollama model
+  and reports tokens-per-second and time-to-first-token, both per-prompt
+  and aggregate. Uses Ollama's streaming `/api/generate` endpoint and the
   server-reported `eval_count` / `eval_duration` for accurate tok/s.
   `--out <file>` writes an `hfo-bench-v1` JSON submission that can be
   contributed to the community leaderboard at
   <https://hfo.carrillo.app/benchmarks/>.
-- **Coding-agent launch manifest + hints.** Every `hfo --launch <agent>`
-  now (1) writes a small JSON manifest to
-  `<configDir>/hfo/agent-launches/<agent>.json` recording which model was
-  bound to which agent and when, and (2) prints a per-agent hint block
-  listing the env vars / config keys the user needs to set for that agent
-  (Claude Code, Codex, Cline, OpenCode, Droid, VS Code, Copilot CLI) to
-  actually route through the local Ollama. hfo never mutates user-owned
-  config files — only records and educates.
-- **AI-search instrumentation**. `docs/llms.txt` (curated site map for
-  ChatGPT / Perplexity / Claude / Gemini) and `docs/llms-full.txt`
-  (flattened prose version of the full documentation).
-- **Five standalone documentation pages.** The homepage sections have been
-  split into real crawlable URLs with unique title / description / canonical
-  / JSON-LD BreadcrumbList:
-  `/install/` (with a `HowTo` schema), `/cli/`, `/keyboard/`, `/privacy/`,
-  `/faq/` (with `FAQPage` schema). Plus `/benchmarks/` for the new bench
-  leaderboard. Sitemap.xml now lists seven URLs instead of one.
-- **Shared sub-page stylesheet** at `docs/assets/page.css` so the new
-  pages stay visually consistent with the landing without duplicating 600
-  lines of hero CSS.
-
-### Changed
-
-- **Portable default install directory.** `hfo` now downloads GGUFs into the
-  current working directory (`process.cwd()`) when neither `--dir/-d` nor
-  `settings.modelDir` is set. The previous behaviour — a hard-coded
-  `D:/Desarrollo/AI/LLMs` (Windows) or `~/AI/LLMs` (Unix) — has been removed.
-  Running `hfo` in any folder now drops its models alongside the project,
-  making the tool fully self-contained.
-- **Install-pill OS detection.** The landing's package-manager switcher
-  auto-adapts its `curl` tab to the visitor's OS — shows
-  `irm … | iex` with the label `powershell` on Windows, `curl … | sh` with
-  the label `curl` everywhere else. Uses `navigator.userAgentData.platform`
-  with a UA-string fallback.
-- **Robots.txt** now disallows `/install.sh` and `/install.ps1` so Google
-  doesn't index raw shell / PowerShell source as text content.
-- **Hero CTA layout**. Removed the `Get started` button (redundant with
-  the pm-tabs above it) and replaced it with `View on npm`. Removed the
-  six trust chips from the hero and moved their content to a quieter
-  data-dense stats strip at the top of the Features section.
-- **Landing hero redesign.** `docs/index.html` now uses a full-viewport
-  (`100vh`) two-column hero with an ambient orb + grid backdrop, a CSS-only
-  package-manager switcher (npm / pnpm / yarn / bun / curl), live trust
-  chips (npm version, CI status, tests, stars, Node ≥ 20, OS support), and
-  a stacked terminal preview that shows scoring bars for the quant picker.
-  Three responsive tiers — desktop ≥ 1025 px, tablet 641–1024 px,
-  mobile ≤ 640 px — replace the previous single mobile breakpoint. Dead
-  rules from the v1 hero were retired.
-- **README hero refresh.** Wordmark is linked, tagline tightened, badges
-  switched to `for-the-badge` style with the brand palette, and the
-  signature terminal snippet matches the site hero for cross-surface
-  consistency. An accidental `claude --resume …` line that had leaked into
-  line 1 of `README.md` was removed.
-
-### Added
-
-- Initial public release. See [README.md](./README.md) for the full feature
-  list.
-- Fullscreen TUI with six tabs (Dashboard, Models, Install, Tune, Help,
-  Settings) on top of a hardware-aware install engine.
-- Headless CLI equivalents for every tab capability:
-  `--view` · `--list` · `--tune` · `--backup` · `--restore` · `--delete` ·
-  `--launch` · `--tab`.
+- Portable default install directory: models download to the current
+  working directory (`process.cwd()`) when neither `--dir/-d` nor
+  `settings.modelDir` is set, making every invocation self-contained.
+- ASCII boot screen that runs while hardware probing / Ollama detection
+  / settings loading happens, and a short "Goodbye" splash on exit.
 - One-liner install scripts for every OS:
   `curl -fsSL https://hfo.carrillo.app/install.sh | sh` and
   `irm https://hfo.carrillo.app/install.ps1 | iex`.
-- Ollama launch-menu shortcut `L` wiring into
-  `claude · cline · codex · copilot · droid · hermes · kimi · opencode · openclaw · pi · vscode`.
-- Zip backup + restore pipeline with level-9 compression and a sidecar
-  `metadata.json` per backup; orphan reinstall that regenerates the
-  Modelfile from the GGUF alone when needed.
-- 20 language packs and 7 themes with a reusable searchable dropdown
-  component.
-- ASCII boot screen that runs while hardware probing / Ollama detection /
-  settings loading happens, and a short "Goodbye" splash on exit.
-- GitHub Pages landing at <https://hfo.carrillo.app> with JSON-LD
-  (WebSite + Person + SoftwareApplication + FAQPage), terminal-styled
-  code blocks, a 404 page, `humans.txt`, `.well-known/security.txt`, and
-  a rasterised favicon + Open Graph pipeline.
+
+### Added — website & AI search
+
+- GitHub Pages landing at <https://hfo.carrillo.app> with full-viewport
+  (`100vh`) two-column hero, ambient orb + grid backdrop, CSS-only
+  package-manager switcher (npm / pnpm / yarn / bun / curl+powershell),
+  stacked terminal preview with scoring bars, and three responsive tiers
+  (desktop ≥ 1025 px, tablet 641–1024 px, mobile ≤ 640 px).
+- Shared `<hfo-terminal>` Web Component at `docs/assets/terminal.js` —
+  every code-block terminal on the site renders through one upgrade point
+  (the hero keeps its unique markup).
+- Install pill detects the visitor's OS and auto-adapts the script tab
+  to `irm … | iex` on Windows and `curl … | sh` on macOS / Linux.
+- Five standalone sub-pages with unique title / description / canonical /
+  JSON-LD BreadcrumbList: `/install/` (with HowTo schema), `/cli/`,
+  `/keyboard/`, `/privacy/`, `/faq/` (with FAQPage schema). Plus
+  `/benchmarks/` for the bench leaderboard. Sitemap lists seven URLs.
+- AI-search instrumentation: `docs/llms.txt` (curated site map for
+  ChatGPT / Perplexity / Claude / Gemini) and `docs/llms-full.txt`
+  (flattened prose version of the full documentation).
+- Stats strip inside Features with live chips (npm version, CI status,
+  tests, GitHub stars, Node ≥ 20, MIT, Windows · macOS · Linux as OS
+  icons). Replaces the noisier trust-row that used to live in the hero.
+- Structured data stack: `WebSite`, `Person`, `SoftwareApplication`,
+  `FAQPage`, `BreadcrumbList`, `HowTo`.
+- Rasterised favicon + Open Graph PNG pipeline (the Pages workflow
+  rasterises `og-image.svg` into `og-image.png` at build time).
+- 404 page with same design system, `noindex, nofollow`, internal
+  recovery links.
+- `humans.txt`, `.well-known/security.txt` (RFC 9116), and a `CNAME` for
+  the custom domain.
+
+### Added — repository hygiene
+
 - GitHub community-health files: issue form templates (bug_report /
   feature_request), pull request template, `CODEOWNERS`,
   `dependabot.yml`, `FUNDING.yml`, `SUPPORT.md`, `SECURITY.md`,
-  `CODE_OF_CONDUCT.md`, `CONTRIBUTING.md`, and this changelog.
+  `CODE_OF_CONDUCT.md`, `CONTRIBUTING.md`, this changelog.
 - GitHub Actions: CI matrix (Ubuntu / macOS / Windows × Node 20 / 22 →
-  typecheck + lint + test + build), Pages deploy (rasterises favicons
-  before publishing), and a release workflow that ships per-OS binaries
-  to GitHub Releases **only** when a `v*.*.*` tag is pushed.
-
-<!--
-When cutting a release, copy the items above into a dated version block
-below, e.g.
+  typecheck + lint + test + build), Pages deploy, and a release workflow
+  that ships per-OS standalone binaries to GitHub Releases **only** when
+  a `v*.*.*` tag is pushed. npm publish is performed manually by the
+  maintainer, not automated.
+- 14 test suites / 96 vitest cases covering every module in `src/core`,
+  `src/infra`, and `src/ui`. Coverage tracked via v8.
 
-## [0.1.0] - 2026-05-01
-### Added
-### Changed
-### Fixed
+### Notes
 
-…and then clear the Unreleased section for the next cycle.
--->
+- Requires Node.js 20 or newer (npm / pnpm / yarn / bun installs). The
+  per-OS standalone binaries shipped with each tagged release have no
+  Node dependency.
+- Ollama is optional: if missing, hfo offers to install it via `winget`
+  (Windows), `brew` (macOS), or the official shell script (Linux).
+- No telemetry, no accounts, no subscriptions — only the two public
+  Hugging Face endpoints and the local Ollama daemon.
 
-[Unreleased]: https://github.com/carrilloapps/hfo/compare/main...HEAD
+[Unreleased]: https://github.com/carrilloapps/hfo/compare/v0.1.0...HEAD
+[0.1.0]: https://github.com/carrilloapps/hfo/releases/tag/v0.1.0