NocoBase 3 review From scaffolding to offline configuration checks
Fixed beta creator, template and CLI tests covering project generation, repeat initialization, database-dialect protection and YAML recovery, with clear Core and Studio boundaries.
AgentIndex · Published 2026-11-04 · Updated 2026-11-04
NocoBase supplies an application foundation intended to let developers and agents concentrate on business-specific logic. The default branch examined here has moved to NocoBase 3, while separate NocoBase 2 documentation remains available. Choosing the version is therefore part of understanding both installation and development practices.
Our review concerns v3 project preparation: whether an official template can be generated, whether configuration states are explicit, whether repeated operations preserve existing files, and whether a configuration error can be recovered from. The results support a scriptable project starting point. They do not establish production readiness of a business application or validate agent collaboration in commercial Studio.
Separate product generations and package versions
The current documentation describes Core as the application foundation and Studio as a commercial workspace for governing people and agents. It also states that v3 packages remain in beta. Product generation, creator version, application CLI version and template version are distinct identifiers. Documentation overview
On November 4, 2026, we used Node.js 26.6.0, an isolated pnpm 11.7.0, @nocobase/create-app 0.1.0-beta.34, default template 1.0.0-beta.87, and @nocobase/app-cli 1.0.0-beta.32. Our repository reference was commit 87f1ea54aebbf25ca0de3481125e6902f1426325. Pinned README
The separately published older @nocobase/cli is not the application CLI tested here. Its package name and command entry point differ. Record the actual package, template and versions when reproducing an installation instead of reducing them to a claim about the latest NocoBase.
What project creation returned
We supplied a project name from its parent directory and used a fixed template with --no-install --json. The response distinguished projectCreated: true, dependenciesInstalled: false and configured: false, and supplied subsequent installation, configuration and development commands. Create an app
| Check | Observed result |
|---|---|
| Project name containing a path separator | INVALID_USAGE; no directory created |
| Fixed official template | Created, including server/runtime.ts |
| Reuse the same project name | Refused; original package manifest unchanged |
| Initialize SQLite configuration | Temporary configuration created |
| Load source with production-only dependencies | Missing TypeScript import mapping reported |
| Supply tsx 4.21.0 explicitly | Offline check passed without findings |
| Initialize again with the same dialect | success-noop; identical config bytes |
| Request PostgreSQL over SQLite configuration | ALREADY_CONFIGURED; file unchanged |
| Malformed YAML | CONFIG_INVALID |
| Restore original configuration | Offline check passed again |
| Machine-readable command catalog | Configuration operations discoverable |
Our reproduction fixture contains the script, versions and outcomes. Initialization generates secrets. The published record retains selected states and error types rather than generated configuration or secret values.
Creation, configuration and operation are different milestones
A directory can exist before dependencies are installed. A configuration file can be written before connectivity, migrations or business routes have been verified. Returning those states separately helps an automation choose its next action instead of treating every successful command as permission to publish the application.
If installation is interrupted, first determine whether scaffolding already completed. That determines whether to retry installation in the existing directory or generate a new project. Our same-name test showed that creation refused to overwrite an existing project. A retry therefore needs to respect the current stage rather than blindly restart the whole sequence.
We generated the official template and ran configuration operations in a disposable copy. No business server was started, no database connection was requested and no migration was executed. Authentication, permission isolation, workflows and data consistency remain separate acceptance areas.
Source mode needs its development environment
To keep the experiment narrow, we installed production dependencies only and skipped installation scripts. That environment ran configuration initialization, but direct loading of TypeScript source could not map a .js import to its .ts implementation. The offline check reported a load failure. Supplying tsx 4.21.0 explicitly allowed the same configuration to pass.
This demonstrates an environment distinction, not a failure of the documented complete installation. Our reduced setup should not be copied as a deployment recipe. Normal development should use the full dependencies required by the selected version. During investigation, record whether commands are operating on source or compiled artifacts. Environment and setup instructions
The official requirements are Node.js 24 or later and pnpm 11, with WSL for Windows. We executed on macOS arm64 only. A successful scaffolding step does not substitute for checking the prerequisites on another host or for qualifying that platform's runtime.
Check repeat-operation behavior explicitly
Repeated initialization with the existing database dialect returned success-noop and preserved the file byte for byte. This is useful for automation because a completed step can be recognized without regenerating secrets merely because the workflow is retried.
Requesting PostgreSQL when SQLite was already configured produced a refusal with ALREADY_CONFIGURED. Those outcomes need different handling: one says the requested state is already present, while the other asks to change the existing state. Treating both simply as commands that wrote no file would discard useful information. Initialization implementation
The malformed-YAML and recovery case is similarly practical. We verified that the error was detected and that restoring the original bytes restored a clean check. This establishes recoverability for that configuration path, not automatic rollback for arbitrary upgrades or database changes.
What remains after an offline check
Every configuration check explicitly used --no-connect. The result therefore does not establish database reachability, correct account privileges or successful migrations. An adopting team should add business-specific cases, such as two identities accessing the same resource, transaction behavior after a failed write, and data retention across restart. Configuration documentation
Component scope matters for licensing as well. The creator and application CLI packages we inspected declare Apache-2.0. That observation is not a blanket license statement for every template, plugin or commercial Studio offering. List the required components and their terms alongside operational requirements. Creator package declaration
For a team wanting agents to work within a consistent application foundation, v3 merits a small project trial. Make repeatable creation and configuration the first acceptance milestone, then test the business behavior actually needed. For operational evidence handling, see Claude Ads; for build-diagnostic evidence, see .NET Skills.