Roadmap

Ciaren is alpha (currently 0.1.x). This page describes the direction, not a dated release schedule. For exactly what's shipped, the code is authoritative and releases are tracked on GitHub.

Where it is today

The core platform works end to end:

• Visual builder with 66 transformation nodes (including 8 chart nodes) plus file, SQL, and cloud-storage I/O
• Multi-engine execution — polars (default) and pandas, selectable per run
• Python code export — readable pandas, polars, or lazy-polars scripts
• Data quality contract nodes (assert not-null/unique/range/expression/row-count)
• Machine Learning (built in) — split, feature engineering, train, predict, evaluate, with MLflow tracking and a Models page
• Scheduling — built-in cron scheduler with retries, catch-up, and auto-disable
• Plugin platform — versioned provider contracts, local + entry-point discovery, and signed .ciarenplugin packaging

The direction

Ciaren is built to become a stable, local-first workflow platform with an extensible ecosystem around it. The plugin provider contracts are the foundation for growing the platform from the outside — without bloating or forking the core.

These are roadmap themes, not promises of specific dated features.

Stable foundation

• Alpha hardening — stabilize the public API, data model, generated code, and behavior of core nodes before moving beyond 0.1.x.
• Portable flow format — formalize .flow files with schema versioning, JSON Schema, migrations, and required plugin/capability declarations.
• Backend source of truth — move node metadata into the backend catalog and have the frontend consume a complete GET /api/catalog/nodes response.

Plugin ecosystem

• Plugin lifecycle — continue improving discovery, loading, permissions, signature verification, install, update, disable, and uninstall flows.
• Community plugin authoring — make it easier to package, test, sign, and document plugins that live outside the open core.
• Production plugin examples — document richer plugin patterns beyond the hello-world node, including tests, packaging, permissions, and code export.

Connectors and data access

• Focused built-in connectors — keep the open core limited to broadly useful database, file, API, and storage integrations while steering long-tail systems to connector plugins.
• Better browsing — improve remote file and table discovery for SQL, object storage, and local-folder connections.
• Connection diagnostics — make test results, permission errors, and setup hints more actionable.

Exporters and portability

• More export targets — explore notebooks, reusable job templates, and other portable artifacts beyond standalone Python scripts.
• Export validation — add checks that generated artifacts can run and match the visual flow behavior.
• Reusable handoff — make exported code easier to version, review, and run in another environment.

Data quality

• Reusable contracts — strengthen validation nodes as first-class data contracts that can be reused across flows.
• Quality reports — surface which checks passed or failed per run, with samples of problematic rows when possible.
• Validation exports — explore exporting quality checks into external test or validation formats.

Machine Learning

• ML workflow maturity — improve metrics, model comparison, lineage, and the Models page.
• MLflow integration — make registered model workflows, aliases, and tracking configuration clearer.
• Guardrails — add stronger warnings for data leakage, risky splits, missing evaluation, and fragile training configurations.
• ML templates — provide starter flows for common classification, regression, clustering, and feature-engineering tasks.

AI capabilities

• AI as an extension point — introduce AI capabilities through providers, not as a required dependency of the core app.
• Assistive workflows — explore pipeline generation, flow debugging, optimization suggestions, and plain-language explanations of errors.
• Data-control safeguards — keep any AI integration explicit about what data is used, where it goes, and how users can opt in or out.

Scheduling and automation

• Schedule observability — improve visibility into upcoming runs, missed runs, retries, failures, and auto-disabled schedules.
• Automation triggers — continue improving REST, CLI, and webhook-based ways to run flows outside the UI.
• Lightweight orchestration — keep scheduling simple enough for local and self-hosted workflows.

User experience and documentation

• Debuggability — improve per-node errors, preview failures, and flow-level troubleshooting.
• Onboarding — add more demo projects, recipes, screenshots, and sample datasets for real workflows.
• Contributor paths — make it clearer when work belongs in core and when it should be built as a plugin, especially for connectors and integrations.

Security and trust

• Plugin trust model — harden permission prompts, signed packages, and manifest validation.
• Operational guidance — document recommended practices for sensitive data, self-hosting, backups, and access control.
• Security review readiness — keep security-sensitive areas testable and auditable as the project matures.

What Ciaren won't become

By design, Ciaren stays focused. It is not aiming to become:

• A real-time streaming engine — batch-style data and ML workflows remain the product center.
• A warehouse-scale orchestrator — heavy production orchestration belongs in dedicated tools; Ciaren should export clean code and integrate with them.
• A multi-tenant SaaS — the core project stays local-first and self-hosted.
• A catalog of every connector — niche integrations should be built and maintained as plugins so the core stays small.
• A black-box runtime — users should be able to inspect, export, and own the code their flows represent.

See How Ciaren Compares for where it fits.

Shape the roadmap

Ciaren is built in the open, and the direction is decided in public — the best way to influence it is to get involved:

• 💬 Discussions — propose features and ideas
• 🐛 Issues — report bugs or focused core/SDK gaps
• 🧩 Build a plugin — extend the platform yourself
• 🤝 Contribute — good first issues are labelled