Frequently asked questions

Answers about data, setup, risk assessment, cost and open-source scope.

On this page

What data leaves my machine?

ArchDev installs local tools and uses hosted services. Requests and shared updates go to ArchDev; model-backed features can send supplied code/context to the selected model provider. Published PR assessments can retain the diff and supporting evidence. A connection alone does not upload your checkout or publish updates. See Your data and access for visibility, retention, providers and revocation.

Do you train models on my data?

The Privacy Policy says Customer Data is not used to train ArchAstro-owned models. Third-party inference follows the applicable provider, contractual terms and technical configuration. Do not assume that every provider route has zero retention. Your data and access explains the distinction.

Which coding agents work with ArchDev?

Use the ArchDev installer on the machine where you run your coding agent, then follow its setup instructions. It is the starting point for Claude Code, Codex, Cursor and other supported coding agents.

Do I need an account?

Yes. Install ArchDev and sign in with your own GitHub account. Organization knowledge and repository features also require the appropriate organization and GitHub access.

Does connecting automatically track my sessions?

No. Your connected agent uses authorized tools to recall context and share approved work updates. Connecting does not start unattended monitoring or publish anything on its own.

How do you define risk?

Risk combines two things. Uncertainty is how much remains unproven about the change working. Consequence is what could happen if it is wrong. A well-tested permission check can still deserve attention because the consequence is high. A fixed rule combines the two component grades; line count does not determine risk. Understand risk and complexity gives the full matrix.

What do you use to calculate risk?

The assessor uses the evidence supplied for that subject and revision: its intent, diff or proposal, relevant source/test context, and available check results and team lessons. Available context differs by assessment path. A model or author judges the components against written definitions; software combines them deterministically. Reasons, coverage and missing inputs matter as much as the label. See Assessment inputs.

How do you know the grade is right?

We compare assessments of frozen changes with researched reference judgments, using deeper code/history investigation and review to identify missing evidence and unclear definitions. This is an evaluation process, not a guarantee that an individual grade is correct or a universal accuracy percentage.

Read the reasons and evidence and challenge a specific claim. If the app offers Is this grade right?, use it to record your view; otherwise send the revision and reason to the team without private code. A disagreement is not proof that a model has already retrained or that a saved grade changed. See Validation and disagreement.

Is a low grade the same as an approval?

No. A grade guides attention, not whether all tests passed or a reviewer approved the change. Not assessed is different from low, and an earlier commit's grade may not apply to the current head. Follow your repository's checks and approval rules before merging.

What does it cost?

Solo is free for individual developers, with no seat subscription. Free access requires an eligible organization. Hosted-model usage is billed separately at provider cost plus 10%, using credit. Eligible organizations receive $20 of model credit on enrollment.

Team costs $49 per seat/month or $492 per seat/year, with hosted-model usage separate. Manage your plan and credit in Settings → Billing. See Pricing and billing.

Is it open source?

Partly. The public ArchAstro/archdev repository contains installers, skills and documentation under the MIT License. That license does not cover its release binaries, which are built from a separate private source repository. The whole hosted product is not open source.

How do I share a finding with my team?

Ask your connected agent to share the finding in the ArchDev stream, confirm its project and approve posting. Other organization members can read it. Keep credentials and sensitive customer data out of the message. Home and shared memory explains how later sessions use the record.

How do my teammates connect?

Share Connect your agent. Each teammate uses their own account. An administrator handles organization and GitHub repository setup; Set up your team walks through both. Everyone uses the same installer; do not share personal credentials.

How is this different from CodeRabbit?

CodeRabbit focuses on code review. ArchDev also connects agent work and team lessons before and after a PR, with approved updates and revision-specific risk evidence. The workflows can sit alongside a separate reviewer. Use ArchDev's grade to decide what to inspect, not as a replacement for a reviewer's findings.

How is this different from Graphite?

Graphite centers on the pull-request workflow. ArchDev adds a shared record of agent activity and lessons across connected sessions, plus pull-request risk assessments. Connecting an agent to ArchDev does not require changing your branch workflow. It is not a promise that ArchDev implements every stacking, merge-queue or review feature of another tool.

Where can I get help?

Contact us. Include the client, account/organization, workflow and non-sensitive error. For a grade, identify the repository, PR and head revision without sending private code. Do not send tokens, passwords or payment card details.