macOS AI writing assistant that works across apps with cloud or local models.
Kerlig is a native macOS AI writing assistant designed to bring model-powered actions into whatever app is already in front of you. The core workflow is simple: select text, invoke Kerlig with a global shortcut, then proofread, rewrite, summarize, translate, draft a reply or run a custom action without moving the work into a browser tab.1 The current product supports macOS 12+ on both Apple Silicon and Intel Macs. The release history lists version 2.8.1, released August 19, 2026, as the latest build; version 2.8 introduced Agentic Mode and MCP-server connections, while 2.8.1 fixed selected-text line-break handling in apps such as Obsidian and Slack.2
The model layer is unusually flexible for a writing assistant. Kerlig says it supports more than 350 models through providers including OpenAI, Anthropic, Google, Groq and OpenRouter, while also supporting Apple Foundation Models, Ollama and LM Studio for local inference.1 This makes the app closer to a system-wide AI client than a single-model grammar checker. Users can switch models, create custom actions, maintain presets and tones of voice, attach documents and images, use a conventional chat interface, edit and regenerate messages, keep chat history and run some actions headlessly in the background.1 Agentic Mode can use approved tools while answering—reading the active window, fetching links, checking the clipboard and searching prior Kerlig chats—and MCP support can connect local or remote servers such as Notion, Linear, Slack, GitHub or custom systems.12
Document handling extends beyond plain selected text. The product can analyze and summarize PDFs, Word documents, ebooks, webpages and images, and its help material documents direct provider integrations using the user's own API keys.1345 Deep Research is also exposed through reasoning/search-capable models rather than through a proprietary Kerlig research engine.16
The main product limitation is platform scope. Kerlig is deliberately Mac-first and has no Windows or mobile client documented on the current official surface. That focus helps the app feel native, but it sharply limits mixed-device teams and users who expect the same writing layer on Windows, iPhone, Android or the web. G2 reviewers repeatedly identify macOS-only availability as the most obvious accessibility constraint.7
Kerlig currently uses a one-time-license model rather than a recurring subscription. The official site lists Basic at $49 for one Mac, Pro at $79 for two Macs and Team at $297 for ten Macs; all three include the core AI features, presets, attachments and support for 350+ cloud and local models.1 The site states that Basic and Pro are currently sold with lifetime updates as a limited-time deal, while the Team tier lists one year of app updates. A 14-day money-back guarantee is also advertised.1
The license price is not the full cloud-usage cost. Kerlig's cloud integrations are bring-your-own-key: users pay OpenAI, Anthropic, Google, OpenRouter or another provider directly for the tokens they consume.1345 Local models through Apple Foundation Models, Ollama or LM Studio can avoid per-token cloud charges, at the cost of using local hardware and managing model setup.1
Older marketplace listings should not be used for current pricing. G2 still shows a $47 Basic figure and Capterra lists $47 Basic and $67 Pro, while the current first-party site is $49/$79/$297.781 Those marketplace prices are useful evidence of historical pricing but are stale for purchase decisions today.
Independent user evidence is positive but not yet large. G2 currently reports 4.9/5 from 19 reviews. Its review synthesis and individual reviews repeatedly praise seamless macOS integration, ease of use, grammar and translation workflows, custom presets and the reduction in browser/app switching. Reported weaknesses include macOS-only availability, the technical burden of configuring individual API keys and requests for more shortcuts or feature depth.7
Capterra reports 5.0/5 from three reviews, with 4.7 for ease of use, features and customer service and 5.0 for value for money. The three-review sample is too small to treat as broad market validation, but the qualitative pattern is consistent with G2: reviewers like the native Mac feel, stability and ability to invoke AI against text in other applications; complaints include incomplete replacement of a full LLM client, missing direct action hotkeys and onboarding friction.8
Product Hunt adds eight reviews at 5.0/5 and 350 followers. Its review summary emphasizes clean UI, ease of use, fast performance, grammar checking, shortcuts and seamless experience.9 Setapp lists version 2.6.0 and an 86% user score from 21 ratings, providing another small but independent Mac-user signal.10 These sources point in the same direction, but the samples are small and likely skew toward early adopters already comfortable with Mac utilities and AI tooling.
Privacy architecture is partly attractive because Kerlig says messages go directly to the selected AI provider rather than through Kerlig servers, and then are stored encrypted on the Mac.1 With on-device Apple Foundation Models or local Ollama/LM Studio models, prompts can remain local.1 That does not make every configuration private by default: when a cloud provider is selected, the text is still transmitted to that provider under its own terms, retention rules and account settings. Users handling confidential material therefore need to choose providers and local/cloud modes deliberately.
No controlled benchmark or broad professional review corpus was located for the exact Kerlig product. The evidence supports claims about workflow satisfaction and product fit more strongly than claims about output quality, because output quality depends heavily on whichever underlying model the user chooses.
Kerlig is strongest for Mac users who write all day across Mail, Slack, browsers, Notion, Figma, VS Code, Jira and similar apps and dislike repeatedly moving text into a separate chatbot.1 The global-shortcut model turns AI into a utility layer over existing applications rather than another destination.
It is particularly attractive to users who already have API access to multiple model providers. Those users can choose a cheap model for routine rewriting, a stronger model for complex drafting, or a local model for privacy-sensitive work without changing the surrounding workflow.1 Researchers and knowledge workers who frequently summarize documents or webpages also benefit from attachments, document chat and Deep Research access.16
The product is less suitable for non-Mac households, teams that require centrally bundled AI usage with no API-key administration, or users who want one vendor to take responsibility for both the application and model inference. G2 reviews make the onboarding tradeoff clear: the same BYOK design that improves provider choice and cost control can be a barrier for less technical users.7
**Strengths** - Native, system-wide macOS workflow reduces copy/paste and browser context switching.17 - Broad support for 350+ cloud models plus Apple Foundation Models, Ollama and LM Studio.1 - One-time license avoids another mandatory software subscription.1 - Custom actions, presets, tones of voice, document attachments and chat make it substantially broader than a basic grammar checker.1 - Current user sentiment is consistently positive across G2, Capterra, Product Hunt and Setapp.78910 - Local-model options can keep sensitive prompts on-device.1 - Version 2.8.1 and an active August 2026 release cadence indicate continuing development.2
**Weaknesses** - macOS only; no equivalent Windows or mobile workflow is documented.17 - BYOK requires users to create, fund and manage provider API keys for cloud use.345 - Cloud usage carries separate token costs in addition to the Kerlig license.1 - Review samples are still small, especially on Capterra and Product Hunt.89 - Output quality is not independently benchmarked and varies with the selected underlying model. - Marketplace pricing can lag current first-party pricing, so buyers should verify the official site before purchase.178
Not separately stated in the source review.
Products matched by shared tasks, audience, workflow, product type and capabilities.