Denker vs Lapu vs Fazm

The products overlap in desktop assistance, but their documented interaction and model-access choices differ. Read each row as a starting point for evaluation, rather than a score or proof that the same task will succeed in all three.

The useful Denker scenario starts with a source you can identify directly: this chart, this email, or this design selection. Pointing or marking supplies the reference; voice or text supplies the requested action and constraints. Denker's documentation also describes tasks, agents, and workspace frames, so you can keep the request connected to a result you review.

A first trial could ask for an unsent reply draft based on an open email, with facts to check before sending. Decide where the draft should appear. Denker's public email guidance leaves sending with the user. For an actual spreadsheet or file change, confirm the supported action and inspect the destination; showing an app is not the same as authorizing its entire account.

Lapu's documentation describes a local desktop agent with file, shell, and accessibility tools. Its model access is built into the product, rather than requiring you to bring an existing Claude Code or Codex login. The floating mini chat provides a goal-entry interface above the apps you use.

That is a meaningful setup distinction if you prefer to manage the app and included reasoning allowance together. Check how its plan accounts for your intended workload and whether the relevant app controls are available. Local files and execution do not imply offline reasoning: Lapu explains that relevant context is sent to its model endpoint through provider APIs.

Fazm documents a native macOS interface for CLI agents, with persistent sessions, chat forking, voice input, and routes into browser and native-app work. It is worth evaluating if you already use Claude Code or Codex and want a graphical session interface that reaches beyond repository files.

Its official repository is MIT licensed and describes a free-to-start path. Verify the requirements for your chosen backend and any browser extension or app access you need. Persistence and forking are organization features; they do not establish that every prior permission, account state, or external write remains usable when you resume work.

Denker Free includes 25 on-screen requests per month, two projects, and five connected tools. Pro is €19 per month with unlimited on-screen requests, projects, and integrations, plus knowledge graph memory. New users receive a 14-day Pro trial without a card or automatic charge. AI provider subscriptions and usage are separate; the included starter allowance is limited.

Lapu's pricing page lists annual Premium at €192, Pro at €576, and Max at €960, displayed as monthly equivalents of €16, €48, and €80. Its FAQ separately describes monthly plans at $20/$60/$100 in the US and €20/€60/€100 elsewhere. These are different billing arrangements: verify your location, checkout, and included allowance before comparing the payment due.

For Fazm, the free/open-source app path does not eliminate upstream provider costs. Compare the additional interface fee, existing provider expense, extra usage, and upfront annual commitment separately. Keep currencies distinct and evaluate the full setup, not only the lowest displayed number.

Separate three questions: which context the agent receives, which actions it can perform, and where processing happens. A local app can still call remote model providers or connected services. Denker uses your chosen CLI and connected tools; Lapu describes its model endpoint; Fazm relies on its configured backends and tools. Review the actual route for your task.

On Mac, screen capture, microphone access, accessibility controls, and browser extensions have different roles. Grant what the chosen workflow requires and check how to interrupt it. This article compares documented controls and does not certify a product's security or data retention. For every tool, review sensitive destinations and external actions at the point they matter.

If the main friction is identifying the exact item on screen, evaluate Denker's selection and handoff. If you prefer a bundled goal-based desktop agent, evaluate Lapu's app controls and included usage. If your work centers on continuing and branching CLI-agent sessions, evaluate Fazm's session interface. These are use-case recommendations from documented design choices, not measured rankings.

Try the same bounded task on two candidates: prepare a reply draft from the same source email, or create a reviewed summary from the same document. Note context supplied, corrections, the destination, and whether the task actually finished. A different tool may fit your second workflow; make the choice from the complete handoff and review.