Browse

The `agent:` resource calls a sibling agent's entire workflow within the same [agency](https://kdeps.com/agencies/) and returns that agent's `apiResponse` output. It is how one agent delegates a subtask to another specialized agent.

`apiResponse:` builds and returns the HTTP response sent back to the caller. It is always the last resource in the dependency chain - the resource pointed to by [`targetActionId`](https://kdeps.com/reference/glossary#targetactionid) in `workflow.yaml`.

The `browser:` resource drives a real browser (Chromium, Firefox, or WebKit) via [Playwright](https://playwright.dev/). Use it to navigate pages, fill forms, run JavaScript, capture screenshots, and maintain authenticated sessions across resources.

The `chat:` resource sends a prompt to a language model and stores the response as the resource's output. The output is the raw response object - the reply text is at `get('id').message.content`. With `jsonResponse: true` the output is the parsed JSON object instead.

Navigate and map a codebase. The `codeIntelligence:` resource has two families

The `component:` resource calls a reusable resource bundle - a registry component installed with `kdeps registry install`, or a custom component in your project's `components/` directory. Think of it like calling a function: you pass typed inputs via `with:`, the component's own resources run, and you get its result back.

The `embedding` executor is a native capability compiled into the `kdeps` binary. It provides a SQLite-backed keyword store for indexing, searching, upserting, and deleting text documents. Use it as the storage layer for RAG pipelines that run fully on-prem.

The `exec:` resource runs a shell command and stores its stdout as the resource's output. Use it for system operations, file manipulation, or wrapping CLI tools that do not have a native resource type.

The `httpClient:` resource makes an outbound HTTP request and stores the parsed response body as its output. JSON responses are parsed automatically; other content types are stored as a string.

The `loader:` resource reads a file, URL, or directory into structured `Document` objects - plain text plus metadata - and optionally splits them into chunks. It is the ingestion step of a RAG pipeline: load, then feed the output into `vectorStore:` or `embedding:` for indexing.

The `python:` resource runs a Python script and stores its stdout (parsed as JSON) as the resource's output.

The `searchLocal` executor walks a local directory and returns matching files by filename glob pattern and/or content keyword. See [Search resources](https://kdeps.com/workflow/resources/search) for how it relates to `searchWeb`.

The `sql:` resource runs SQL queries against a named connection and returns the result set as the resource's output. Use it to read, write, or transact against any supported database.

The `telephony:` resource models in-call actions (answer, say, ask, menu, dial, record, hangup, and more) for Twilio-compatible telephony providers. The provider POSTs its call webhook to your kdeps API route; the resource reads the webhook fields from the request body and builds a TwiML response that is returned via the standard `apiResponse` mechanism.

The `vectorStore:` resource adds documents to, and runs similarity search against, an external vector database. Unlike `embedding:` (SQLite keyword/vector index built into kdeps), `vectorStore:` talks to a real vector database service - Qdrant, Chroma, Pinecone, pgvector, and more - for production-scale RAG.