Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM/Return Editor Review: Local-First AI Contract Review for Mac and Windows
RAG & Document Chat

Return Editor Review: Local-First AI Contract Review for Mac and Windows

·7 min read·By Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI
Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

Return Editor (version 1.4.1 as of 10 October 2026) is a closed-source desktop app for Mac and Windows that reviews contracts with AI, and its free plan does so with a local model that, according to the developer, keeps documents on your computer. It marks risky clauses, answers questions with clause citations, and drafts fixes that you approve one by one. Cloud AI is a paid, opt-in extra, and most claims here come from the product's own pages.

Return Editor is a desktop document editor for Mac and Windows that reviews contracts with AI: it flags risky clauses, answers questions with citations to the clause they come from, and proposes rewrites that you accept or reject. Its Free plan runs a bundled local model with no account, and a paid Pro plan adds optional cloud AI. The app is closed source. The developer submitted it to the directory and disclosed that they build it; PromptQuorum checked the details against returneditor.ai and its changelog on 10 October 2026 and has not tested the app hands-on.

Return Editor Review: Local-First AI Contract Review for Mac and Windows

Key Takeaways

  • Developer: a sole proprietorship registered in Kraków, Poland, according to its privacy policy; the person who submitted the listing is the developer.
  • Workflow: Analyze flags clauses, Ask answers with citations, Fix proposes rewrites, and Returns re-runs saved checks such as key-date extraction.
  • Local engine: a bundled llama.cpp server that runs downloadable models; the default model is a 5.3 GB download and 16 GB of memory is recommended.
  • Plans: Free (local, no account) and Pro (adds cloud AI); a firm-oriented Counsel tier is announced but waitlist-only.

📍 In One Sentence

Return Editor uses a local or cloud language model to flag risky contract clauses, answer questions with citations, and suggest rewrites, and its code is closed.

💬 In Plain Terms

You open a contract, the app highlights clauses that may be risky, and you ask questions about it; on the Free plan the model runs on your own computer.

📌Note: This review is based on returneditor.ai pages checked 10 October 2026, plus a listing submission from the developer. PromptQuorum has not tested the app or the accuracy of its risk flags, and AI output is not legal advice.

What Is Return Editor?

Return Editor is a Markdown-based document editor with a contract-review workflow built in: you import a contract, and the app analyzes it, answers questions about it and rewrites passages in place. The developer's submission describes it as a Tauri and Rust desktop app that bundles a llama.cpp server for local inference.

The directory lists it under Contract review and Document & PDF chat: it can answer questions about any imported document, but its main output is clause-level findings and rewrites rather than open-ended conversation.

Get It

Return Editor is a free download for Apple Silicon Macs (macOS 14 or later) and 64-bit Windows 10/11; Linux has a notification list only.

Item
Where to get it
MacDownload page: notarized .dmg, Apple Silicon only
WindowsDownload page: signed installer, no admin rights
LinuxNot available; email notification list
Websitereturneditor.ai
Privacy policyPrivacy policy
TermsTerms
ChangelogChangelog
Source codeNot published

Version as verified on 10 October 2026: 1.4.1, from the changelog. Windows SmartScreen may show a "Windows protected your PC" warning for the newly signed installer; the download page says to choose More info, then Run anyway.

Plans and Pricing

The Free plan needs no account and has no cap on the local AI; the Pro plan adds cloud AI for a monthly or yearly fee.

Plan
Price
What it adds
Free$0Local AI, local search, no account
Pro$39/month or $390/yearCloud AI, 500 actions/month
CounselNot yet availableWaitlist; audit log, lockdown mode

Prices as listed on returneditor.ai at the time of checking; they may change, and the Counsel tier is announced, not released.

Features Confirmed by the Sources

Every item below comes from the product's own pages; none has been independently tested.

  • Analyze. Imports PDF, DOCX, ODT, RTF and HTML, converts them to Markdown, and marks findings in traffic-light colors: red for risk, amber for items that need attention.
  • Ask. Answers questions about a document with citations to the relevant clause, and runs semantic search across a whole folder of Markdown files.
  • Fix. Proposes inline rewrites that you accept or reject; nothing changes until you choose.
  • Returns. Saved, reusable checks, such as key-date extraction, that you run on any document; the documentation describes chaining several model passes.
  • Editor. Variables and formulas inside documents, Mermaid charts, and export to PDF and HTML with charts and images carried over since version 1.3.6.

Local vs. Cloud AI

In the Free plan the AI is a bundled llama.cpp engine that runs models on your own machine and works offline once a model is downloaded. The home page lists Qwen3.5 9B, Granite 4.1 8B, GPT-OSS 20B and Bielik 11B v3, a Polish-specialist model.

Cloud AI is a Pro option: requests go through the maker's proxy, hosted by Hetzner in Germany, to Anthropic, which according to the privacy policy keeps them for up to 7 days and does not train on them.

  • Hardware. The default model is a 5.3 GB download; the site recommends 16 GB of memory.
  • Idle behavior. The local model unloads after 30 minutes without a question by default; the changelog says this can be changed in settings.
  • Images. The local model cannot read images yet, while cloud AI can read images in chat and in documents.
  • Telemetry. The privacy policy says the desktop app has no client-side telemetry; the free app contacts the maker's server only for update checks (which can be disabled), crash reports you consent to, and support requests you send.

📌Note: These are the maker's declarations, not audit results, and the code is closed. PromptQuorum has not inspected network traffic; for confidential contracts, check the mode shown in the status bar and test with the network disconnected.

Trade-Offs: Benefits vs. Limitations

Local AI on the Free plan

What it means in real use:
Contracts stay on your computer; works offline.
Limitation / caveat:
Needs a multi-GB model download and enough RAM.

Cited answers

What it means in real use:
Each answer points to the clause it came from.
Limitation / caveat:
Citation accuracy was not tested here.

Reviewable rewrites

What it means in real use:
Fixes are proposals you accept or reject.
Limitation / caveat:
Quality depends on the model you run.

Mac and Windows

What it means in real use:
One app on both, signed and notarized.
Limitation / caveat:
Apple Silicon only on Mac; no Linux build.

Cloud option

What it means in real use:
A stronger model and image reading on Pro.
Limitation / caveat:
Content leaves your machine for Anthropic.

Who Should Use It

  • People who review contracts, policies or long reports and want the first pass to stay on their own computer. The Free plan is built for this.
  • Small teams that want clause-level flags and cited answers without a per-seat cloud service. Cloud AI is optional.
  • Polish-language readers. The model list includes Bielik, a Polish-specialist model.

What We Could Not Verify

  • Accuracy of the risk flags. PromptQuorum did not run the app, so which clauses it flags, misses or mis-rates is not assessed.
  • What the flags are tuned for. The pages we read do not say which contract types, jurisdictions or standards the analysis is based on.
  • Source code and audits. No public repository, third-party audit or certification was found, so privacy statements cannot be checked against code.
  • Company and install base. The site names no team; the privacy policy identifies a sole proprietorship in Kraków, Poland, and customer numbers are not published.
  • Legal reliability. AI-generated flags and rewrites are not legal advice; a qualified lawyer should check any contract you sign.

Competitors and Alternatives

PrivateGPT

Platforms:
macOS, Windows, Linux
Price / license:
Free / Apache 2.0
Key difference:
Offline RAG over your files, developer-oriented
Articles about PrivateGPT (8)

Also mentioned in:

+7 more not shown

HilbertRaum

Platforms:
Windows, macOS, Linux
Price / license:
Free / GPL-3.0
Key difference:
Portable offline workspace with document Q&A
Articles about HilbertRaum (2)

Sidekick

Platforms:
macOS
Price / license:
Free / MIT
Key difference:
Local chat over your files, Mac only
Articles about Sidekick (3)

Competitor details change often; confirm each app's current price, license, and platforms on its own listing.

Compare in directory

Frequently Asked Questions

Is Return Editor a replacement for a lawyer?

No. The app flags and rewrites with a language model, and models can miss or misread clauses; treat its output as a first pass and have a qualified lawyer review anything binding.

Is Return Editor open source?

No. The developer's submission lists it as proprietary and closed source, and no public source repository was found.

Do I need an account?

Not for the Free plan. Pro needs an account, because cloud requests are tied to an account for billing and rate limiting.

Does it work in languages other than English?

The pages we read do not list interface languages. Search uses a multilingual embedding model (bge-m3) and the model list includes the Polish-specialist Bielik, so test your own language on the Free plan first.

How often is it updated?

The changelog shows frequent releases: Windows support arrived in 1.3.4, and the latest release, 1.4.1, is dated 9 October 2026.

Verdict

Return Editor combines three things that are usually separate: clause-level contract review, cited document Q&A and a free on-device model, in one Mac and Windows app. Against that, it is closed source, young (Windows support dates from version 1.3.4), cloud AI sends content to Anthropic, and nothing here has been tested hands-on, including the quality of its risk flags. It suits people who want a private first pass over their own contracts; readers who need auditable code can compare AnythingLLM or PrivateGPT.

Sources

  • Return Editor — features, plans and local and cloud modes, checked 10 October 2026.
  • Download page — platforms, installers, version and signing notes.
  • Changelog — version 1.4.1, release dates and recent features.
  • Privacy policy and security page — data flows, retention and hosting.
  • Documentation — editor and AI feature overview.
  • Listing submission from the developer, received by email, who disclosed building the app.

← Back to Power Local LLM