[Feature]: Workflow & Automation Integration (n8n / Agent Pipelines)

[sarthakdubeyy12]

Description

Add support for workflow-based integration so Moss can be used as a retrieval component inside automated pipelines. This could include a lightweight adapter or community node that exposes core actions (index documents, query search, manage data) as callable steps within a workflow. The goal is to allow developers to trigger Moss operations programmatically and chain them with other services (LLMs, storage, notifications) without writing custom glue code, making it easier to use Moss in end-to-end AI systems.

Use case

Moss is great for fast semantic search inside applications, but integrating it into real-world workflows currently requires manual setup and custom glue code. In many use cases (like support systems, document search, or AI pipelines), developers need to trigger indexing, run queries, and pass results to other systems automatically. This is especially useful for teams building end-to-end AI workflows, not just standalone apps.

Proposed solution

Provide integration support for workflow and agent orchestration tools (e.g., an n8n community node or similar adapter). This would allow users to trigger indexing and querying as part of a workflow, and pass results downstream (to LLMs, notifications, databases, etc.). Adding a few example pipelines would make it easier for developers to use Moss in real production scenarios without writing custom integration code.

Alternatives considered

No response

[yatharthk2]

Looks interesting to me, are you proposing this as a package integration or a cookbook ?

[yatharthk2]

also can you please provide me surface peudocode on how the integration would look like and what features would it support.

[sarthakdubeyy12]

Looks interesting to me, are you proposing this as a package integration or a cookbook ?

This would be a package integration first.

The idea is to make Moss usable as a reusable component inside workflows, not just provide examples. By exposing core actions like indexing and querying as callable steps, developers can plug Moss directly into automation pipelines without writing custom glue code each time. also we can still add a few cookbook examples later after this, but the main focus is on building a lightweight integration layer.

[yatharthk2]

lets do one thing, you can start with creating a package but if its just a thin wrapper on top of moss client, it might not qualify as a package because it poses no advantage to the community. We will then convert it to a cookbook, should be fairly easy.

[sarthakdubeyy12]

also can you please provide me surface peudocode on how the integration would look like and what features would it support.

simple view of how the integration could look -

Step 1: index or load data
await mossNode.execute({
action: "create_index",
input: { index: "support-docs", documents }
});

Step 2: query
const result = await mossNode.execute({
action: "query",
input: { index: "support-docs", query: "refund time", topK: 3 }
});

Step 3: pass structured output
if (result.success) {
nextStep(result.data);
} else {
handleError(result.error);
}

as how the integration would work -

• Provide a lightweight adapter that exposes core Moss operations as callable steps
• Each step takes input from the previous step and returns structured output
• Designed to plug into workflow/orchestration tools (like n8n) and chain with other services

initial feature scope -

• Index creation / loading
• Document ingestion (add / upsert)
• Semantic query (topK)
• Metadata filtering
• Structured JSON output for downstream steps

Notes on scope and usage -

• This would act as a thin wrapper over the existing SDK, not replace it
• Intended for workflow and pipeline use cases (automation, agent orchestration), not latency critical in app queries
• Keeps Moss as the core retrieval engine while the workflow tool handles orchestration

The integration would act as a thin wrapper on top of the existing SDK, exposing core actions like indexing and querying as workflow steps. This keeps Moss focused as a fast retrieval engine, while the workflow layer handles orchestration.

[sarthakdubeyy12]

lets do one thing, you can start with creating a package but if its just a thin wrapper on top of moss client, it might not qualify as a package because it poses no advantage to the community. We will then convert it to a cookbook, should be fairly easy.

Got it, i will start with a small package and evaluate the value. If it ends up being too lightweight, I will convert it into a cookbook.

[SiddhantSwan12]

Hi! I'd like to work on this. Planning to build an n8n community node that wraps the Moss TypeScript SDK, exposing query, ingest documents, delete documents, and create index as n8n node operations — following the same cookbook pattern as the existing examples/cookbook/langchain/ and examples/cookbook/dspy/ examples. Will add it under examples/cookbook/n8n/ with a README and a sample workflow JSON. Can you assign me?

[sarthakdubeyy12]

Hi! I'd like to work on this. Planning to build an n8n community node that wraps the Moss TypeScript SDK, exposing query, ingest documents, delete documents, and create index as n8n node operations — following the same cookbook pattern as the existing examples/cookbook/langchain/ and examples/cookbook/dspy/ examples. Will add it under examples/cookbook/n8n/ with a README and a sample workflow JSON. Can you assign me?

Thanks for jumping in, I have already started working on a prototype from my side based on this idea.

[SiddhantSwan12]

That's awesome! Would love to collaborate on this which parts are you planning to cover? Happy to pick up whatever's left.