DocKit puts an agent and five NoSQL clients in one Tauri desktop app

One window for five different databases

The product in this repository is described in two ways, and the first one is a little surprising. The repository is dockit, the package is named dockit, and the download page is under dockit, while the README heading and the product copy call the app DocKit. The naming is not consistent across the artefacts, so if you are looking for this tool in a package manager or an issue tracker, search for both.

What the app does is not ambiguous. It is a native desktop GUI client for Elasticsearch, OpenSearch, DynamoDB and MongoDB, plus EasySearch, and the positioning is to replace browser consoles like Kibana and the AWS Console with a single native desktop app. The description in the package manifest adds a fifth engine, and lists the features the app is known for: an agentic desktop GUI with natural language queries, visual management and monitoring, and a privacy-first stance with your data and your keys kept on your machine.

The audience is a developer or operator who works with more than one of these backends. That is the niche, and it is a real one. The alternative to this app is two or three browser tabs with different authentication flows, different query languages, different result formats and no way to keep a saved query in one place. The unified surface is the product.

The technical shape is a Tauri application. The badges list macOS, Windows and Linux, the Tauri framework and Rust, and the repository makes that concrete with a src-tauri/ directory for the Rust side and a src/ directory for the TypeScript front end built with Vite, with UnoCSS for styling and Vue components. The private package manifest, version 1.4.5, declares the type as module and lists Tauri plugins for the clipboard, dialogs, filesystem access, global shortcuts, the operating system and process control, which is the plugin set you would expect from a desktop client that reads files and talks to remote services.

The agent writes the query, and asks before it deletes

The Agentic Data Studio is the feature that distinguishes the app from a plain query GUI, and it is described as: describe what you need in natural language and the agent writes queries, inspects schemas, updates documents, deletes records, creates indexes and returns results.

The specifics matter. Query generation covers Elasticsearch DSL, PartiQL and MongoDB queries, so the agent is not translating into one dialect. Schema inspection means the agent reads your schema and explains it. Data operations cover CRUD, index management and bulk actions through conversation, which is a large surface for a tool that can act rather than only suggest.

That is why the safety section is the most important part of the description. The agent operates with per-source permissions, there is a confirmation gate for destructive operations, and credentials are never exposed to the LLM. The third of those is a design claim worth testing rather than trusting, but the idea is right: a query generator only needs your schema, not your password, so a client that keeps the key in the OS keychain and hands the model a redacted connection is meaningfully safer than one that pastes a connection string into a prompt.

The supported model providers are OpenAI, Anthropic, DeepSeek, Requesty, Opper, Atlas Cloud, Ollama and LM Studio, and the app is explicit that you bring your own key. The Ollama and LM Studio entries mean a fully local path exists, which is what makes the privacy-first claim coherent rather than decorative: with a local model and a local keychain, nothing about your schema leaves the machine.

Per-database depth is where the real work is

The bulk of the feature documentation is per engine, and it is where you can judge whether the app covers your actual cases.

For DynamoDB there is a query builder with primary key filtering and more than 13 operators, a PartiQL editor with autocomplete, syntax highlighting and inline item CRUD, and table management covering creation, modification, indexes for global and local secondary indexes, TTL, streams, point in time recovery and encryption. DynamoDB Local is supported for offline development without AWS credentials, and the authentication options are AWS profiles, SSO, access keys and IAM roles.

For MongoDB there is a query editor with autocomplete and bulk write support, a document browser with pagination and inline CRUD, a manage view for indexes, storage statistics and collection metadata, connection settings including authentication, TLS and replica set configurations, and import and export in JSON, CSV and JSONL.

For Elasticsearch and OpenSearch there are separate connection types with independent configurations, so the two do not share a credential store. The Monaco editor, which is the engine behind VS Code, supports JSON5, inline comments and field autocomplete from the live mapping. The cluster management covers node health, shard state, index tracking and alias control, and there is a grammar catalog of more than 120 API endpoints for reference. The stated version support is Elasticsearch 1.x through 9.x and OpenSearch 1.x through 3.x, which is a wide range to maintain and a good sign of how much of the API surface the app knows.

The Common and Ultimate split runs through all of this. The free tier includes every data source, unlimited connections with basic direct authentication using user and password or static access keys, TLS, OS keychain credential encryption, the full Monaco editors and visual query builders, browsing and inline editing, result set export to CSV, JSON, Markdown or the clipboard, query history and saved queries, metadata browsing, and DynamoDB Local. The paid tier adds the AI features, cluster management and monitoring, a built in MCP server, batch and file level import and export, remote connection enhancements such as SSH tunnels and HTTP or SOCKS proxying, and cloud credentials for AWS profiles, SSO and IAM.

One build, two modes, and what the licence actually covers

The licensing model is the part to understand before installing, and the project states it directly: DocKit follows a single official build with a runtime Community or Ultimate mode, while the source remains Apache 2.0.

So there is one binary, and it decides at runtime whether you are in the free mode or the paid one. The free mode is described as free forever and covers everything except the Ultimate list. The Ultimate tier is priced at 9.9 dollars a month or 99 dollars a year, with a seven day free trial, and the features it unlocks are the agent, natural language to SQL, explain, optimize and one click fix with your own LLM key, cluster management and monitoring, the built in MCP server, batch and file level import and export, the remote connection enhancements and the cloud credential handling.

Two rules about the subscription are worth reading closely. A version lock says that every release published while a subscription is active stays unlocked forever, offline, even after the subscription ends, and the seven day trial unlocks everything during the trial but does not grant version lock rights. And cloud services, with sync named as an example, require an active subscription.

What the Apache 2.0 source licence does and does not do is worth separating. The code is open, so you can read it, build it and modify it, and the paid tier is a runtime switch rather than a closed component you cannot inspect. What the licence does not give you is the hosted service, and the version lock mechanism means that a build you make yourself is your own business. Whether the paid features in the source would enable you to build your own equivalent is the sort of question to answer by reading the code, and the agent directory in the repository is where that code lives.

The practical summary is that this is an open source project with a commercial product built on it, sold by the same team, and the free tier is generous enough to be a complete tool for direct connection and editing work.

How the repository is developed and released

The engineering setup in the manifest tells you how seriously the project takes its own quality, and the Tauri split makes the test story unusual for a front end project.

The test script runs the JavaScript suite and the Rust suite together:

"test": "jest --runInBand --coverage --coverageReporters json-summary text html lcov && cargo test --manifest-path src-tauri/Cargo.toml",
"build:macos": "tauri build -t universal-apple-darwin"

So a single command produces Jest coverage across the front end and runs the Rust tests in the Tauri crate, and the CI variant runs the JavaScript half with the ci flag. The build script type checks with vue-tsc before running the Vite build, the lint scripts are ESLint with a fix variant, and the macOS build produces a universal binary for both Apple silicon and Intel.

The repository layout supports that story. There is a tests/ directory, a jest.config.cjs, a codecov.yml for coverage reporting, an eslint config, a.husky/ directory for git hooks, a.prettierrc, an uno.config.ts for the utility CSS setup, an auto-imports.d.ts and components.d.ts generated by the tooling, an AGENTS.md, a CHANGELOG.md, a CONTRIBUTION.md, a docs/ directory, scripts/ and public/, and an.omo/ directory whose purpose the documentation does not explain.

The release history shows regular work on the 1.4 line: v1.4.3 on 2026-08-20, v1.4.4 on 2026-09-03 and v1.4.5 on 2026-09-06, with the manifest at 1.4.5. The last push was on 2026-10-18 and the repository is not archived, so the work is current, and the presence of a Chinese README alongside the English one suggests the project is maintained with a wider audience in mind.

Editorial conclusion

DocKit suits a developer who works across more than one NoSQL backend and wants a single native window with consistent editors, saved queries and local credentials instead of a browser console per vendor, and it suits anyone who wants the agent's natural language queries without sending data to a hosted service. It is the wrong choice if you need the agent and cluster monitoring on a licence, because those are the paid tier, and a poor match if you expect the free build to be a complete product, because the paid features are gated by a runtime mode rather than by a separate build. Before adopting, check the Ultimate feature list against what you actually need, note that cloud services such as sync require an active subscription, and install from the official download or release artefact rather than assuming a package repository carries it.

Frequently asked questions

Official sources

1. geek-fun/dockit on GitHub
2. License: Apache-2.0
3. Project website
4. README
5. Releases