Roadmap question: is visual workflow orchestration planned for the Agent module?

[jxdang]

Hi WeKnora team,

Thanks for open-sourcing the project — the modular RAG pipeline and the "knowledge-as-tools" design of the Agent mode are genuinely well thought out, and we are evaluating WeKnora for an internal enterprise knowledge scenario (self-hosted, Docker, v0.8.3).

I'd like to ask about the roadmap for the Agent module:

Are there plans to add a visual workflow orchestration / builder capability in upcoming releases — similar to Dify's Workflow canvas (node-based graph with LLM, retrieval, condition, loop, code and tool nodes) or MaxKB's workflow editor?

Context on why we're asking: today a single ReAct loop isn't enough for our business processes. We need to encode deterministic multi-step flows, for example:

1. Classify an incoming document by type and route it to different processing branches.
2. Retrieve from different knowledge bases per branch, then merge and rerank.
3. Call an external API / MCP tool, validate the payload, retry or fall back on failure.
4. Insert a human-in-the-loop approval at a specific step, then resume.
5. Produce a structured, reproducible and auditable output instead of letting the LLM choose the next step each turn.

A node-based canvas would also let non-engineers on our team assemble and maintain these flows, which the current agent configuration doesn't really allow.

Specifically, we'd love to know:

• Is workflow orchestration already on the roadmap? If so, which release are you targeting, and is there an RFC or design doc we can follow?
• If planned, what form will it take — a visual canvas (Dify/MaxKB style), a declarative DSL (YAML/code pipeline), or both (a visual editor that compiles to a definition file)?
• Will workflow steps be able to reuse existing capabilities as first-class nodes — KB retrieval, chunk/graph/wiki tools, MCP tools, Agent Skills, sandbox/browser, long-term memory?
• Will each run be persisted and traceable at step level (per-node inputs/outputs) for debugging and audit?
• Is there an extension point available today we could use as a workaround — e.g. chaining agents through the REST API, or composing the pipeline via the built-in MCP server / weknora CLI?

Even a rough direction would help us a lot with planning. Happy to share more detail on our scenario, or to help test an early design.

Thanks!