Dari Docs
A self-deployed CLI that has coding agents attempt real developer tasks from your documentation and propose targeted edits.
At a Glance
Publicly downloadable tool deployed using your own infrastructure and model-provider keys. No software purchase or subscription is listed. Model and sandbox usage are billed separately to your accounts. No software license is displayed in the repository.
Engagement
Available On
Alternatives
Listed Nov 2026
About Dari Docs
Dari Docs, from the mupt-ai organization, is a CLI that tests whether documentation works by having a coding agent attempt real developer tasks from it. It reports blockers and, in optimize mode, passes the findings to an editor agent that proposes updated files. The latest published release is v0.1.6.
What It Is
Dari Docs is a documentation testing and improvement tool built on the Harbor framework. A tester agent reads selected local docs or public URLs and tries a task, such as installing an SDK and making a first API call, in a fresh sandbox. The output is feedback on what it tried, what worked, and where it got stuck. It is not a docs-quality score or an automatic merge.
How the Workflow Runs
The check command runs the tester only. The optimize command also sends aggregate feedback and the original docs to an editor, which writes proposed files to .dari-docs/updated/ along with a changelog. Several tasks can be run in parallel and compared across models, producing individual reports and an aggregate-feedback.md. Source docs stay untouched unless --apply is passed. Public URLs, including llms.txt manifests, are supported for check only.
Harnesses, Models, and Sandboxes
Claude Code, Codex, Pi, and OpenCode have built-in Harbor adapters, and custom adapters can be configured. The bundled deployment uses Modal, and other Harbor sandboxes are configurable. The tester and editor can use different harness and model settings. Compatibility depends on the combination of harness, model, and sandbox. A runner exposes an async API for submitting runs, checking status, reading events and results, and cancelling.
Setup Path
The tool is self-deployed rather than hosted. It requires uv, Python 3.13+, a Modal account, and a model-provider key. The README notes that the Harbor runner is not in the latest published release, so users build the CLI from source with Go. dari-docs init extracts the editable runner, prompts, and skills, which are then deployed to Modal. Live verification with product credentials is opt-in.
Community Discussions
Be the first to start a conversation about Dari Docs
Share your experience with Dari Docs, ask questions, or help others learn from your insights.
Pricing
Self-Deployed
Publicly downloadable tool deployed using your own infrastructure and model-provider keys. No software purchase or subscription is listed. Model and sandbox usage are billed separately to your accounts. No software license is displayed in the repository.
- Agent-based documentation testing
- Proposed documentation fixes
- Configurable agent harnesses, models and sandboxes
- Batch tasks with configurable parallelism
- CLI and self-deployed runner
Capabilities
Key Features
- Agents attempt real developer tasks from docs in fresh sandboxes
- Feedback reports and aggregate-feedback.md
- Optimize mode proposes edited files and a changelog
- Choice of harness, model, and sandbox
- Separate tester and editor settings
- Batch tasks and model comparison with parallelism
- Check published docs via public URLs and llms.txt
- Async runs API with status, events, results, and cancellation
- Opt-in live verification with product credentials
- Review before apply with --apply flag
