Discover the Best AI Tools

Turns typed text into editable event cards, checklists, timers, calculators and other small utilities. Works offline with keywords by default; optional Jev classification sends input to an API, while saved cards stay in browser storage.

Voice-driven macOS app and browser automation using local speech transcription and remote Jev decisions. Requires Apple Silicon, system permissions and API credentials; command transcripts leave the machine, and optional text models add external calls.

Self-hosted read-only access to personal balances, transactions, liabilities and investments via Plaid. Requires your own Plaid credentials and bank linking; deployments are single-user and need private access or authentication.

Local chat-shaped command bar uses Jev to select MCP tools and arguments, then renders replies from returned data. Requires a TypeSafe key and any tool-specific credentials; it is a local development setup and classification or source errors remain possible.

Experimental Android overlay reads visible chats, uses Jev to score context and a generative model to draft ranked replies. Requires Accessibility permissions and API access; it fills but does not send, Lark body capture is incomplete, and intent scores are speculative.

CLI and MCP client submits and tracks Forma benefit claims, with optional AI extraction from receipts. Requires a Forma login and an inference provider for AI features; receipts may be sent externally and generated claims need review before submission.

Apple Silicon Mac workspace manages model downloads, local inference and AI chat, voice and media applications, with optional cloud providers. Capabilities depend on installed engines, model licenses and hardware; cloud features require credentials and are not local-only.

Local web tool analyzes pasted relationship chats for emotion, intent and reply suggestions using Jev or an LLM. Requires provider access and sends chat content for analysis; scores cannot establish another person’s feelings or intentions.

Unofficial MCP server queries Copilot Money’s local Mac cache and optionally enables live reads or write tools. Cache-only reads stay local, but connected AI providers receive returned financial data and enabled writes modify the account.

MCP integration exposes T-Bank accounts, documents, transfers and shopping or travel workflows to assistants. Requires local banking authentication and sessions; these tools can perform real financial actions and depend on changing mobile-bank interfaces.

MCP connector sends bank-statement PDFs to MainBook for extraction and balance checks, returning JSON, Excel or CSV. Requires an account and page credits; processing is remote and flagged transactions still need review.

DocJev classifies and splits PDF, DOCX and PPTX files using local LiteParse text extraction and hosted Jev decisions, with optional cloud OCR. Office files need LibreOffice. Splits are contiguous page ranges; oversized inputs can fail explicitly. The local demo is not an offline inference service, and small recorded benchmarks do not establish broad accuracy.

Finlynq manages personal transactions, budgets, portfolios and imports, with AI chat and read/write MCP tools. It offers managed hosting or Docker self-hosting. Its bearer API keys grant unscoped account access until regenerated; the supported integration is MCP, while REST response shapes may change. Modified hosted versions carry AGPL obligations.

Desktop suite for editing documents, spreadsheets, slides and PDFs, with AI assistance and CLI automation. Files can stay local, but AI requests use the chosen provider; packaged builds enable optional usage analytics by default.

Validates structured electronic invoices using the hosted Eleata service, with local format and error-code reference tools. Validation requires an API key and uploads invoice content; free usage is quota-limited.

macOS voice orb combines local Whisper transcription with Jev choices over Accessibility controls. Build the current source and provide a TypeSafe key and permissions; inaccessible or pixel-only interfaces are unsupported, and broad goals need human verification.

Unofficial read-only MCP connector unifies multiple Mercury banking organizations using a token per organization. Structured fields are filtered, but free text and explicitly enabled PDF downloads can expose sensitive banking data to the AI client.

Otto is an early macOS/Windows desktop assistant using accessibility and OCR, with reviewed Jev actions or model-free exact form filling. Optional planning uses a separate model; guided runs need provider keys. Broad app compatibility and trusted release signing remain unfinished, and native-control tests do not establish general model task success.

Jev Skip reads YouTube captions, asks Jev to score transcript segments and can skip likely sponsor sections. It needs a paid TypeSafe key and sends transcript text externally. No captions means no detection; false skips occur, Chrome is the tested browser, and published small-fixture results used a gateway rather than the direct API.

Native Mac automation combines local visual/accessibility perception with remote Jev action selection; it needs Apple silicon, macOS permissions and a provider token, sends observed text externally, and does not generate typed text or plan arbitrary long workflows.

Chrome extension controls the current browser session from DOM snapshots using Jev and a separate text helper. Requires model credentials and sends page context externally; it lacks iframe and shadow-root support and needs supervision on signed-in sites.

Experimental Chromium extension combines an FX planner with Jev action selection for browsing and extraction. Requires model-gateway access and a compatible browser; page context goes to remote models, and complex frames or widgets can fail.

Experimental Apple Silicon Mac app turns voice or text requests into Accessibility actions using Jev. Requires macOS permissions and a model API key; speech is English-focused and there is no confirmation queue before actions.

Computer-control agent combines visual planning with hardware or software mouse and keyboard control. Requires model access and device setup; the optional Safari Jev mode uses DOM data and remains experimental.