Cai is live · free · last verified 26 Oct 2026 · viability monitor
Cai
Free, open-source AI action layer for macOS that runs custom actions on any selection with ⌥C
Cai earns a place in a Mac dock because it does one thing that matters: it turns a selection into a finished action without leaving the window. The MCP authoring flow — Claude Code, Cursor, or Codex proposes over MCP, you review a diff with capability chips, approve, and it then runs locally from ⌥C — is the part most competitors haven't copied, and it makes the $0 MIT-licensed price feel almost unfair. Budget an afternoon to wire up your first action chain. If you want a conversational assistant with memory, or you're not on a Mac, look at ChatGPT or Raycast instead.
Verdict generatedVerified 4d ago · liveness 68/100 · cite: rightaichoice.com/tools/cai
- macOS power users who want a keyboard-triggered AI action layer without leaving the current app
- Developers filing GitHub issues or Linear tickets straight from an error log or screenshot
- Privacy-minded users who want prompts processed on-device with no account or telemetry
- People turning repetitive text work — translation, summarization, grammar — into one-shortcut chains
- Windows and Linux users — Cai is a macOS 14+ menu bar app
- People who want a conversational chatbot with memory and chat history
- Intel Mac owners expecting the built-in Ministral 3B model to run without pointing Cai at Ollama, LM Studio, or a cloud
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Cai if you want a conversational assistant with chat history, or if you're not on macOS 14+ — it is a keyboard-triggered action layer for selections, not a chatbot, and Apple Intelligence support needs macOS 26+.
Cloud model usage is billed by the provider, not Cai — connecting Anthropic or OpenRouter means their API key and their per-token charges apply to every action run through it
Cai is $0: MIT-licensed, no subscription, no account, and no per-run token cost on the built-in Ministral 3B model. That puts it below paid launcher subscriptions like Raycast Pro and below per-seat AI writing tools. Your only variable cost is optional cloud inference you wire up yourself through OpenRouter, Anthropic, or an OpenAI-compatible endpoint, billed by that provider.
In short
Cai — Free, open-source AI action layer for macOS that runs custom actions on any selection with ⌥C. Best for macOS power users who want a keyboard-triggered AI action layer without leaving the current app, Developers filing GitHub issues or Linear tickets straight from an error log or screenshot, Privacy-minded users who want prompts processed on-device with no account or telemetry. Free to use.
What is Cai?
Cai — Free, open-source AI action layer for macOS that runs custom actions on any selection with ⌥C. Best for macOS power users who want a keyboard-triggered AI action layer without leaving the current app, Developers filing GitHub issues or Linear tickets straight from an error log or screenshot, Privacy-minded users who want prompts processed on-device with no account or telemetry. Free to use.
Cai is a free, MIT-licensed macOS menu bar app that turns whatever you select into an action. Select text or an image, press ⌥C, and Cai runs an action you built or that your coding agent proposed: an AI prompt, a shell script, a URL template, OCR, or a connector call that files a GitHub issue or Linear ticket. It reads the selection and surfaces likely actions for the content type — an error log offers "create ticket," a screenshot offers "image to text," a meeting invite offers a calendar event. You never leave the app you were in. The design is local-first. Cai ships with a built-in local model (Ministral 3B via MLX) that downloads automatically, so prompts run on-device with no API key, no account, and nothing per-run. Apple Intelligence joined as a local provider on macOS 26+ (announced July 2026), alongside LM Studio, Ollama, the HuggingFace MLX community, and opt-in cloud endpoints such as Anthropic and OpenRouter. Anything that touches the network does so because you wired it up. Named secrets live in the macOS Keychain and reach a model only as {{secrets.NAME}} placeholders. Two features separate it from a clipboard utility. Action chains save a multi-step pipeline — translate then replace, or fetch a URL, summarize, and send to Slack — and trigger the whole thing from one shortcut. The MCP server makes Cai a target for coding agents: Claude Code, Cursor, or Codex can write an action, but it arrives as a reviewable diff with capability chips, and you approve it before it runs. Version 1.6.0 (August 2026) refined that diff approval flow and sped up the built-in model. The audience is narrow and specific: people on macOS 14+ who work from the keyboard and would rather not paste into a chatbot. Raycast covers launcher and extension territory; Cai's angle is agent-authored actions plus a local default, at zero cost.
Key features
- Press ⌥C on any selected text or image to run an action in place
- Built-in local AI model (Ministral 3B via MLX) downloads automatically and runs offline
- Apple Intelligence as an on-device model provider on macOS 26+
- Action chains pipe a selection through AI, shell scripts, and destinations in one keystroke
- MCP server lets Claude Code, Cursor, and Codex author actions as reviewable diffs
- Approve-before-run diff review with capability chips flagging what each action can do
- Image to Text (OCR) from screenshots or copied images, then chain further actions
- Create GitHub issues and Linear tickets directly from a selection
- Shell actions run any CLI tool: gh, kubectl, jq, ffmpeg
- Destinations include webhooks, AppleScript, deeplinks, and Apple Shortcuts
- Bring your own model via LM Studio, Ollama, HuggingFace MLX models, Anthropic, OpenRouter, or any OpenAI-compatible endpoint
- Named secrets stored in macOS Keychain and exposed to models only as {{secrets.NAME}}
- Clipboard history with search, pin, and reuse for the last 100 items
- Per-app context awareness surfaces different actions depending on the active app
- Keyboard-first navigation with favorites pinned to ⌘1–9 and type-to-filter
Behind the verdict
Cai's best idea is that the selection is the interface. You don't open anything, you don't paste anything, you don't pick a model from a dropdown. Press ⌥C on an error log and the actions that make sense for an error log are already there; press it on a screenshot and OCR is on the list. Per-app context sharpens this further — a Reply action in Mail drafts something different from the same action in Slack. That is real workflow engineering, not a prompt wrapper. The local default is the second thing it gets right. Ministral 3B via MLX downloads automatically, so the app is useful the moment you install it with brew install --cask cai, and the FAQ's claim that prompt actions default to the built-in local model means your ordinary summarize/translate/grammar runs cost nothing per invocation. Apple Intelligence on macOS 26+ adds a second on-device option without you supplying a key. Cloud providers — Anthropic, OpenRouter, anything OpenAI-compatible, or a local server via Ollama or LM Studio — are strictly opt-in, and secrets referenced as {{secrets.NAME}} never appear to the model. Strengths beyond those two: action chains genuinely compose. Selection → fetch article → summarize → send to Discord is one keystroke, as is stack trace → extract error + repro → create Linear ticket → Slack ping. Shell actions mean any CLI you already have (gh, kubectl, jq, ffmpeg) becomes a Cai action without an integration being written for it. GitHub and Linear are built-in connectors; everything else routes through webhooks, AppleScript, deeplinks, or Apple Shortcuts. The clipboard history — 100 items, searchable, pinnable — covers the case where you needed a clipboard manager after all. Weaknesses are mostly about who this is not for. It is macOS 14+ only, with Apple Intelligence gated to macOS 26+. Intel Mac owners can run the app but should expect to point it at Ollama, LM Studio, or a cloud key rather than lean on the built-in model. It is not a chatbot: there is no conversation memory or chat history, and if that is what you want, Cai is the wrong shape of tool. The flexibility is also an obligation — you edit prompts, URLs, and shell scripts, because Cai surfaces actions rather than shipping a pre-built library for every task. And there is no centralized admin, SSO, or multi-user management, so a company trying to standardize automation across a department will hit a wall. The honest positioning: Cai is one of the few tools where "free and open source" is a description of the architecture, not a discount. It runs on your machine, it keeps working when the vendor does nothing, and the MIT license means you can inspect the action execution path yourself. Against Raycast it trades ecosystem breadth for a local default and agent-authored actions; against a clipboard manager like Maccy it trades passive storage for active execution. Pick it for the 20 minutes of copy-paste drudgery it deletes from every day.
Models under the hood
as of 2026-10-24
Limitations
- Cai is a macOS-only menu bar app requiring macOS 14+, and Apple Intelligence as a provider specifically requires macOS 26+.
- It defaults to a small built-in model (Ministral 3B via MLX), so heavier reasoning tasks will push you toward Ollama, LM Studio, or a cloud key you supply yourself.
- Agent-authored actions are proposal-only: every action must be approved as a reviewable diff before it runs, which is a safety feature but also a deliberate friction in the agent workflow.
- There is no chat interface, no conversation memory, and no centralized admin or SSO for organizations.
- Setup is not zero for real use — you edit prompts, URL templates, and shell scripts to get past the built-in actions.
as of 2026-10-01
Researching Cai? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is Cai still active?
Cai is live · free · last verified 26 Oct 2026 · viability monitor.
Viability score
How well maintained and how widely used is Cai? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →What's new in Cai
Checked 10 days agoAcross the latest 2 updates: 2 feature updates.
Cai v1.6.0 released
Introduced the latest version with improved agent-authored actions, refined diff approval flow, and performance optimizations for the built-in local model.
Apple Intelligence support added on macOS 26+
Cai now supports using Apple Intelligence as a local model provider on macOS 26+, giving users another on-device option beyond the built-in Ministral 3B.
What do users say about Cai?
What people actually say about Cai — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
ResearchedDiscussed by users across 4 sources (Hacker News, Product Hunt, Stack Overflow, Lemmy) · researched Sep 22, 2026.
Weighted by the 59 posts each of 4 sources contributed.
- +⌥C next to ⌘C builds fast muscle memory and skips window-switching entirely
- +Local-first design with zero telemetry and no account echoes real user priorities
- +MIT license and open source mean users can audit or fork the app
- +Bundled Ministral 3B via MLX gives offline AI with no setup and no per-run tokens
- +Works as a standalone app after the developer's Raycast extension stalled in review
- −Barely any post-launch user reports means reliability and edge cases remain unverified
- −Built-in local model requires Apple Silicon, excluding Intel Mac owners entirely
- −MCP agent workflow and diff-approval claims have no community usage evidence yet
- −Favorites bound to ⌘1–9 conflict with common app shortcuts, per user complaints
- −Single maintainer raises real continuity and long-term support risk
- • Cloud model usage via OpenRouter or Anthropic is billed by those providers
- • No paid tier exists, so there is no commercial support contract to buy
How we verified this
Every listed fact about Cai— pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 26 Oct 2026. Pages mix model-assisted synthesis with human review; see how we verify.
Verification history
We have re-verified Cai 10 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed SSL certificate
- — changed limitations, models, switching in, pricing fit, switching out, who should skip it, integrations, our verdict, setup time, features, pricing tiers, use cases, who it suits, hidden costs, description
- — changed models, limitations
- — changed summary, who should skip it, pricing tiers, our verdict, features, integrations, description, who it suits
- — changed limitations, models
- — changed domain expiry
Showing the 6 most recent of 10 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does Cai cost in 2026?
Cai: Free to use. Pricing last verified 26 Oct 2026.
Plans compared
Every published Cai tier, its price as the vendor publishes it, who it fits, and what it adds over the tier below. The 12-month cost projection follows.
| Plan | Price | Billing | Who it's for |
|---|---|---|---|
| Free & Open Source | Free | — | Mac users on macOS 14+ who want inline AI actions and agent-authored automation without a subscription, account, or per-run token billAdds: Starting tier — the only tier: MIT-licensed app, built-in Ministral 3B local model, unlimited custom actions and chains, MCP authoring with diff approval, and GitHub/Linear connectors, all at $0 |
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where Cai's pricing actually pencils out — and where peers do it cheaper.
Cai is $0: MIT-licensed, no subscription, no account, and no per-run token cost on the built-in Ministral 3B model. That puts it below paid launcher subscriptions like Raycast Pro and below per-seat AI writing tools. Your only variable cost is optional cloud inference you wire up yourself through OpenRouter, Anthropic, or an OpenAI-compatible endpoint, billed by that provider.
Has Cai's pricing changed?
No plan price for Cai has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick Cai?
Pick Cai if you are macOS power users who want a keyboard-triggered AI action layer without leaving the current app, Developers filing GitHub issues or Linear tickets straight from an error log or screenshot, Privacy-minded users who want prompts processed on-device with no account or telemetry, People turning repetitive text work — translation, summarization, grammar — into one-shortcut chains.
Look elsewhere if you are Windows and Linux users — Cai is a macOS 14+ menu bar app, People who want a conversational chatbot with memory and chat history, Intel Mac owners expecting the built-in Ministral 3B model to run without pointing Cai at Ollama, LM Studio, or a cloud, Organizations needing centralized admin, SSO, or multi-user management.
Skip it if Skip Cai if you want a conversational assistant with chat history, or if you're not on macOS 14+ — it is a keyboard-triggered action layer for selections, not a chatbot, and Apple Intelligence support needs macOS 26+.
GeneratedReal-world workflow fit
Concrete scenarios for the personas Cai actually fits — and what changes day-one when you adopt it.
A stack trace is sitting in a terminal window. Select it, press ⌥C, run the chain 'stack trace to ticket': extract error + repro, create a Linear ticket, then ping Slack.
Outcome: A filed, readable ticket and a Slack notification in one keystroke, without leaving the terminal or retyping the error.
Select a German paragraph in a document and run the saved 'Translate in place' chain: selection → Translate to Spanish → Replace Selection, on the built-in local model.
Outcome: The selection is replaced in place with no app switch, no paste, and no token cost.
Connect the agent from Settings → Connections → Agents, describe the action you want over MCP, review the proposed diff with its capability chips, approve it, then run it from ⌥C with the agent out of the loop.
Outcome: A custom action exists that the agent authored but that you vetted, and it runs locally with no agent and no tokens involved.
Use cases
- Press ⌥C on a Slack error message to create a GitHub issue with AI-summarized context
- Select a paragraph and run 'Translate to German' with automatic replacement of the selection
- OCR a screenshot of a meeting invite and pipe the text to create a calendar event
- Trigger a shell script that renames files based on selected text, then send confirmation to Slack
- Highlight code and run 'Explain this code' on the built-in local model without leaving your editor
- Chain selection → fetch article → summarize → send to Discord in one keystroke
- Select an address and hand it to Maps without retyping it
- Have Claude Code or Cursor propose an action over MCP, review the diff, then run it from ⌥C
Setup time & first value
How long it actually takes to get something useful out of Cai — broken out by persona, not the marketing-page minute.
Install via Homebrew (brew tap cai-layer/cai && brew install --cask cai) and you can press ⌥C within minutes — the built-in Ministral 3B model downloads automatically, so no key or account is needed for prompt actions. Plan roughly an afternoon to wire up custom actions, connectors, and your first action chain. Expect a little longer if you are pointing Cai at Ollama, LM Studio, or a cloud
Switching to or from Cai
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From Raycast or Spotlight: keep both installed — Cai acts on the current selection rather than launching apps, so the two cover different keystrokes
- →From Maccy or another clipboard manager: import nothing; Cai's clipboard history holds your last 100 items, but the value is acting on a selection instead of storing it
- →From ChatGPT or another chat app for text tasks: recreate your common prompts as Cai AI prompt actions so they run inline from ⌥C
- →From a personal collection of shell aliases and Automator workflows: wrap each one as a Cai shell action or action chain so it fires from the selection
- ↗To Raycast: rebuild text and AI actions as Raycast extensions if you need a launcher and a broader extension ecosystem more than a selection-first shortcut
- ↗To ChatGPT or Claude: move to a chat app if you want conversation memory and back-and-forth refinement rather than fire-and-forget actions
- ↗To Apple Shortcuts alone: replicate simple chains natively if you only need a handful of automations and want to avoid installing another menu bar app
- ↗To Zapier or Make: move server-side if your automations need to run without your Mac awake or need to be shared across a team
Integrations
Resources & guides
Tutorials & Learning
YouTube returned 6 videos for “Cai”, and we withheld 6: 6 could not be judged, because “Cai” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Cai.
Official links
What are the best Cai alternatives?
The closest alternatives to Cai we list are Raycast, WPS Office and Flot.ai. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| Raycast | Freemium | Pricing not published | — |
| WPS Office | Freemium | Pricing not published | — |
| Flot.ai | Freemium | Pricing not published | — |
Alternatives to Cai
View allRaycast
Raycast is a keyboard-first Mac launcher with an AI agent that commands your apps, files, and tools from one hotkey.
WPS Office
WPS Office for iOS is a free AI office suite that edits Docs, Sheets, Slides, and PDFs — with AI slide generation, PDF signing, and
Tools that pair well with Cai
Common stack mates teams adopt alongside Cai, with the specific reason each pairing earns its keep.
Raycast
Raycast is a keyboard-first Mac launcher with an AI agent that commands your apps, files, and tools from one hotkey.
WPS Office
WPS Office for iOS is a free AI office suite that edits Docs, Sheets, Slides, and PDFs — with AI slide generation, PDF signing, and
Flot.ai
Flot.ai puts an AI writing and reading copilot in a floating sidebar over any macOS or Windows app — plus an AI Memory that recalls what
Cai vs its closest rival
Cai's closest head-to-head is with Truleo — Choose Cai if you're a macOS power user wanting a free, private, keyboard-driven AI assistant that automates text actions locally — ideal for devs and productivity enthusiasts. Choose Truleo only if you're a law enforcement agency needing to surface leads across siloed data like jail calls, RMS, and BWC — its niche capabilities are unmatched but irrelevant outside LE. Verdict generated
Head-to-head comparisons
Cai vs Truleo
Choose Cai if you're a macOS power user wanting a free, private, keyboard-driven AI assistant that automates text actions locally — ideal for devs and productivity enthusiasts. Choose Truleo only if you're a law enforcement agency needing to surface leads across siloed data like jail calls, RMS, and BWC — its niche capabilities are unmatched but irrelevant outside LE.
GeneratedCai vs Locus Robotics
Cai and Locus Robotics serve completely different purposes: Cai is a free, open-source macOS tool for local AI actions on text/images, ideal for individual power users; Locus Robotics is an enterprise AMR solution for warehouses, requiring a RaaS subscription. Choose Cai if you're a macOS user wanting privacy-focused automation on selections; choose Locus if you operate a high-volume fulfillment center needing 2-3x productivity gains.
GeneratedCai vs Presto Voice
Cai and Presto Voice serve entirely different needs. Cai is a free, local, keyboard-driven AI assistant for macOS power users who want privacy and automation in text/image tasks. Presto Voice is a contact-priced enterprise voice AI platform for QSR chains to automate drive-thru ordering and boost revenue via upselling. Choose Cai for personal productivity, Presto Voice for restaurant operations.
GeneratedFrequently asked questions
Categories
Best-of guides
Used Cai? Help shape our editorial sentiment research.