Updated on
PaellaDoc: reviews and analysis
The autonomous local software factory: from discovery to validated code on your Mac.
Our own productPaellaDoc is built by this site's author. It is scored with the same axes as every other tool.
Affiliate link · no extra cost to you
Our verdict
PaellaDoc is the autonomous agent that works on your machine instead of the cloud: a local software factory connecting discovery, specs, planning, AI execution and validation as one system with product memory. Against cloud autonomous agents like Devin, here your data, repository and evidence never leave your Mac, and the models (Claude, Codex, Gemini) work inside its validation gates. Free for individual use.
Best for: Mac developers building products with agents who lose context between sessions.
What changed this week
Sweep of
Second week with no dated third-party conversation: neither the vendor forum (last activity 14 June) nor GitHub Discussions (since April 2025) have moved. The only observable change is a higher build number, from 0.2.390 to 0.2.408, within the same 0.2 alpha series and with no dated changelog to back it as a product fact. All five axes stay put.
- down
- no change
We have tracked it since July 25, 2026: 3 sweeps on record, with the score moving between 3.7 and 3.9. We publish the latest change here; the full series is not published.
What the internet says
A week with no verifiable news. Every corner where third-party conversation might live got checked again, GitHub Discussions, the vendor's own forum, Hacker News, Reddit, Product Hunt, X, and none of them show activity dated this week: the vendor forum has not moved since 14 June and GitHub Discussions since April 2025. The only observable change is a higher build number on the download, from 0.2.390 to 0.2.408, but it stays in the same 0.2 alpha series, with no dated changelog and no new signing on Windows or Linux. July's licence commitment, free up to three projects, remains the last word on price.
What the web repeats in favour
- Free up to three projects with no account needed, and data kept locally in plain SQLite
- It does not tie you to one model: it works with Claude, Codex, Gemini or local engines, and logs the real cost of every call
- The plugin SDK lives in its own repository and its licence file is Apache-2.0, so the extensible part can be audited and extended
- The licence promise published in July puts in writing that the current product stays free for individual use, with the three-project limit
What the web repeats against
- There is almost no public conversation and no independent reviews, which leaves a buyer without outside contrast
- The only discussion involving users other than the author sits on the vendor's own forum, running on the vendor's own infrastructure, with several replies signed by its bots
- The only signed and notarised build is the macOS Apple Silicon one: the Windows and Linux ones ship unsigned and as invite-only alpha
- It remains at alpha 0.2 and a Pro tier exists with no figure published anywhere, so anyone outgrowing three projects cannot know today what they will pay
- The framework's historical repository has been archived and read-only since 19 May 2026, under AGPL-3.0 plus a commercial licence
Sweep sources: GitHub · GitHub · Official pricing · Docs · Communities · Communities
Pros / Cons
Pros
- Full autonomous cycle: from idea to validated code
- Everything local: data, repo and evidence on your Mac
- Persistent product memory across sessions and models
Cons
- macOS with Apple Silicon only
- Needs third-party agents to execute the code
TLDR: PaellaDoc is the autonomous software factory that works on your Mac instead of the cloud: it connects discovery, specifications, planning, AI execution and validation as one system with persistent product memory. It does not replace the models (Claude, Codex, Gemini): it orchestrates them inside validation gates with evidence for every step. Free for individual use, with an inspectable core and all data local.
What PaellaDoc is and how it works
Coding agents have a problem none of them solves alone: they execute very well and forget between sessions. The product decision you made on Tuesday, the reason an architecture was discarded, the acceptance criterion you defined at the start: all of it lives in scattered conversations the next agent never sees.
PaellaDoc attacks that problem by becoming the factory where agents work: a local macOS application that keeps product discovery, specifications, planning, AI work, code and proof connected as one living system. Every task a model executes inherits the product’s full context and leaves verifiable evidence of what it did.
The second design decision is locality: data, repositories and evidence never leave your machine. Against cloud autonomous agents, here the whole factory lives on your Mac, with an inspectable core and an extension SDK under an open license.
What it is like day to day
The change from using loose agents shows in the second session: you never explain the product again. The context that a normal flow rebuilds by pasting files and repeating decisions persists here, and the agent executing today knows what was decided yesterday and why.
The full flow walks the product cycle: you capture discovery (what is being built and for whom), specifications derive from it, planning slices the work and the models execute each piece inside validation gates that check the result before accepting it. That structure is what makes delegation calm: the agent does not advance without passing the gates, and every step leaves a trail.
An interesting side effect of that structure: given durable context and validation, small cheap models perform above expectations, because they work on a bounded problem instead of guessing. The obvious friction is the platform: macOS with Apple Silicon only, and anyone on Windows or Linux is left out.
Pricing and plans
Individual use is free, up to three projects and no account required, with a declared promise that personal use stays free. The core is source-available and the extension SDK carries an Apache license. Data is portable and inspectable, on your machine.
The economics carry one important nuance: PaellaDoc does not include the models. Execution is done by Claude, Codex or Gemini with your existing subscriptions or keys, so your factory’s real cost is what you already pay for your agents. What the tool adds, the memory, orchestration and validation layer, currently costs nothing for individual use.
Who it is for (and who it is not for)
PaellaDoc is for developers and product builders on Mac who already work with agents and have hit their real limit: context lost between sessions and the lack of structure when delegating. The longer the project, the more persistent memory pays against the paste-context-every-conversation flow.
It is also the natural option for anyone who cannot or will not let their code pass through cloud factories: the comparison with Devin is exactly that, local autonomy with your data at home versus remote autonomy as a service.
It is not for anyone working outside macOS, nor for anyone who only wants editor autocomplete: without the intent to delegate the full cycle, the factory is oversized and a standalone agent like Claude Code covers the case.
PaellaDoc alternatives
Devin is the direct counterpoint: the same autonomous engineer concept, run as a cloud service on subscription, against the free local factory. To orchestrate several agents in parallel without the product layer, Conductor solves that specific piece. And as the executor inside the factory or on its own, Claude Code is the reference. They are in PaellaDoc alternatives and the full ranking in the best AI coding tools.
Frequently Asked Questions
Does it replace Claude Code or Codex?
No: it orchestrates them. PaellaDoc supplies the durable context, the planning and the validation gates, and the models from Anthropic, OpenAI or Google execute inside that structure. You need your subscriptions or keys for those models.
What does local actually mean?
That the application, the data, the repositories and the evidence live on your Mac. There is no cloud service your code passes through, and what gets sent to the models is what each specific task needs.
Is it really free?
For individual use, yes: up to three projects, no account, with the declared promise that personal use stays free. The core is source-available and the extension SDK is Apache-licensed.
Does it work on Windows or Linux?
No: it is a macOS application optimized for Apple Silicon. Outside the Mac, the alternatives run through cloud autonomous agents or composing your own flow with standalone agents.