PatentForge v0.5.0 — merged-product release, cloud or local

[scottconverse]

PatentForge v0.5.0 is the first release of the merged product. PatentForgeLocal (the local-only fork) and the original PatentForge (the cloud-only product) are now one application called PatentForge. Pick Local mode (Ollama + Gemma 4 on your hardware) or Cloud mode (Anthropic Claude via your own API key). Switch any time in Settings.

This release closes an 8-run merge plan that took the two prior products and stitched them together behind a single LLM provider abstraction.

What's new

• Two modes, one product. Local mode (free, fully offline, runs on your hardware via Ollama + Gemma 4) or Cloud mode (Anthropic Claude via your own API key, pay-per-call against your own Anthropic account). Same prompts, same pipelines, same outputs — your choice where the model runs.
• Two installer editions. Lean (no Ollama bundle, cloud-only, ~200 MB) and Full (Ollama + Gemma 4 bundled, ~880 MB, supports both modes). Pick at download time based on whether you want the option to run locally.
• Provider chooser in Settings. First section on the Settings page. Local panel (Ollama URL + local model) and Cloud panel (encrypted Anthropic API key + cloud model). Switching providers preserves both sides' settings.
• First-run wizard branches. Full installs open with a Local/Cloud chooser; Lean installs skip straight to Cloud setup.
• Cost-confirm modal before every Cloud-mode run. Estimated USD cost shown; Approve required before any Anthropic API call is made. Local mode bypasses the modal entirely (it's free).
• Provider-aware cost displays. Sidebar totals, per-stage cards, and tab estimates read "Free" in Local mode and $N.NNN in Cloud mode.
• Edition-aware tray. The Go system tray spawns Ollama as a child process only when the install edition is Full AND the user's provider is Local. Cloud-mode users on Full installs don't run Ollama (RAM save).

What's the same

• Six-service federated architecture (NestJS backend + Express feasibility + 3 Python services + frontend served by backend).
• 6-stage feasibility analysis with live SSE streaming, prior art search, claim drafting, compliance checking, and full application generation.
• USPTO PatentSearch + Ollama Cloud Web Search optional integrations.
• The same exports (Markdown / Word / HTML / CSV) with the same legal disclaimers.
• All API keys (Anthropic, USPTO, Ollama Cloud Web Search) encrypted at rest using your machine's keychain-derived key.
• Same legal posture — research tool, not legal advice.

Migration from PatentForgeLocal

If you have PatentForgeLocal v0.1.x installed: install the new Full edition over it.

• Your existing data, settings, and downloaded Gemma 4 model are preserved.
• Provider defaults to Local. Your install behaves exactly as before.
• The SQLite database file patentforgelocal.db is silently renamed to patentforge.db on first boot. Your projects carry over.
• Cloud mode is now available as the first section in Settings if you want to try it. Bring your own Anthropic API key.

Migration from the original PatentForge cloud-only product

If you have the original scottconverse/patentforge clone: that repo is being archived. The merged product replaces it.

• Install the new PatentForge (any edition) and set provider to Cloud in Settings.
• Paste your Anthropic API key into the Cloud panel.
• Your old project data is in the original repo's database; export-import is a manual step right now (issue tracked).

Behind the scenes (the 8-run merge plan)

For the curious — here's how we got here. Each run shipped behind audit + a PR.

• Run 1 — Parity audit between the two repos. docs/parity-audit.md produced.
• Run 2 — LiteLLM provider abstraction across all four services (3 Python + Node feasibility). LLMClient boundary, retry pushed to LiteLLM, per-service dispatch.
• Run 4 — Backend AppSettings.provider + encrypted cloudApiKey columns; idempotent migration; SQLite CHECK constraint enforcing 'LOCAL' | 'CLOUD'.
• Run 5 — Frontend Provider chooser in Settings with conditional reveal panels. Model pricing tables for both Anthropic and Ollama.
• Run 6 — Installer split (Lean / Full), tray edition-aware service management, FirstRunWizard branching by edition, CostConfirmModal wired into ProjectDetail. Plus the Run 5.5 follow-ups (provider-aware cost rendering across the UI).
• Run 7 — Docs rewrite covering both modes. README, USER-MANUAL, ARCHITECTURE, landing page, GitHub Discussion seed. Frontend in-app branding cleanup.
• Run 8 (this release) — Version bump to v0.5.0 across every surface, silent DB-rename startup hook, CHANGELOG [Unreleased] → [0.5.0], service-side branding cleanup tail, release.

Verification

841 automated tests green across all subprojects:

• backend Jest: 340/340 (+11 db-migrate tests this run)
• frontend Vitest: 231/231
• tray Go test + vet: green
• 3 Python pytest suites: 89 + 92 + 71 = 252
• feasibility npm test: 29/29
• docker compose config + 3 installer scripts: clean

Cleanroom end-to-end verification (full live stack with real Ollama + Anthropic API) is an operator pre-cutover step, not part of the autonomous test suite.

Known follow-ups

These are tracked as focused sub-issues post-v0.5.0:

1. Node feasibility service Cloud branch — currently throws LLMClientCloudNotImplementedError pending the Anthropic streaming + tool-call adapter. The 3 Python services already work in both modes via LiteLLM; running feasibility in Cloud mode is queued.
2. ollamaWebSearchApiKey column restore — Run 4 dropped ollamaApiKey on a misreading of the field's purpose. Frontend retains the input field; backend column needs to be restored under a clearer name.
3. 5 pre-existing tsc errors — ClaimsTab.tsx:428 ClaimData/ClaimNode mismatch + 4× useViewInit.test.ts InventionInput fixture rows missing projectId. Baseline carried since pre-merge.
4. FirstRunWizard chooser as <button> rather than <input type="radio"> — slight a11y polish surfaced in Run 6 audit Lens 2.
5. Binary filename rebrand (patentforgelocal-tray.exe → patentforge-tray.exe, etc.) — coordinated with the build pipeline; post-release follow-up.
6. Go module path rebrand (github.com/scottconverse/patentforgelocal/* → .../patentforge/*) — coordinated with the GitHub repo rename.
7. GitHub repo rename (scottconverse/patentforgelocal → scottconverse/patentforge) and archive of the upstream scottconverse/patentforge cloud-only repo — operator-driven destructive actions executed post-merge.

Where to download

GitHub Releases — pick the Lean or Full edition for your platform:

• Windows: PatentForge-{Lean,Full}-0.5.0-Setup.exe
• macOS: PatentForge-{Lean,Full}-0.5.0.dmg
• Linux: PatentForge-{Lean,Full}-0.5.0.AppImage

(Release artifacts produced by the per-platform installer build scripts post-merge.)

Credits

PatentForge stands on excellent open-source work:

• Ollama — local LLM runtime
• Google Gemma 4 — open-weight language model
• Anthropic Claude — frontier cloud LLM
• LiteLLM — unified provider abstraction
• context-mode — context window compression

Legal note

PatentForge is a research tool, not a legal service. The author is not a lawyer. The AI is not a lawyer. AI-generated analysis may contain errors, omissions, or fabricated references. Always consult a registered patent attorney before making filing, licensing, or enforcement decisions.

See LEGAL_NOTICE.md for the full disclaimer.

Have questions? Open an issue, or post in this Discussions thread.