nele.ai
AIReach chat completions across multiple models through nele.ai, the German company AI assistant gateway, with one API key. Agents run generation under a data posture German compliance teams already accept.
What This Integration Enables
nele.ai is built for the German company that wants one AI gateway rather than a per-team scramble, and its most interesting design choice is not the model list. It is that a document collection carries an explicit list of model IDs that are permitted to query it. That turns a knowledge base into something closer to an access-controlled asset, which is exactly the shape a people operations or legal team needs when the documents in question are contracts, works agreements and personnel policy. The vendor positions the platform for German companies and their compliance expectations; FlowRunner's contribution is making sure the permission list stays a decision rather than a default.
Practically, an agent gets the whole surface behind one API key. It can create and manage knowledge databases, then add documents either by uploading them or by having nele.ai fetch them directly from a protected URL with basic, bearer or custom-header authentication. It embeds them for search and runs semantic search that returns snippets with their source filenames and any context stored alongside them. Chat completions can be grounded in a collection or given live web search, but not both in the same call, which is a constraint worth designing around rather than discovering at runtime. Image generation and audio transcription round out the same connection. Everything an agent does here is a step in a flow that a person can inspect, which is the difference between a knowledge base and a black box.
Without FlowRunner
With FlowRunner
Use Case Scenarios
A policy knowledge base that maintains itself
Documents are published to an internal store. The agent calls Add Document From URL with the credentials that reach the protected source, letting nele.ai fetch the file itself rather than shuttling bytes through the flow, and enables auto embed so the document is searchable immediately. List Document Collection Items confirms the index finished. When someone asks a policy question in Slack, the agent runs Search Document Collection, grounds Create Chat Completion in the returned snippets, and answers with the source filenames attached. The staff member gets the answer and the document it came from, in the same message.
Recorded sessions turned into searchable material
A training session or works council meeting is recorded. The agent runs Transcribe Audio against the file, which nele.ai handles through Azure Whisper or Azure Fast Transcription and accepts up to 500 MB, so a full session does not need splitting. Create Chat Completion summarises the transcript into decisions and owners, the summary is filed in Confluence, and the transcript itself is added to the relevant collection so future questions can reach it. Passing the source language on the transcription call is a small step that pays for itself on multilingual sessions.
Onboarding answers scoped to the joiner
A new hire starts in Personio or BambooHR. The agent resolves their department and location, picks the collection scoped to that group, and answers their first-week questions grounded in it. Upload Image Attachment lets the joiner send a photo of a form they are stuck on and get an answer about that specific field. Get User Settings runs first as a cheap check that the document collections module is enabled for the account, so the flow fails at setup rather than in front of a new employee.
Human-in-Loop Highlight
The consequential operation in this connector is not a completion. It is Update Document Collection changing the permitted model list. A collection holding works agreements and personnel files is readable by exactly the models named on it, and adding an ID to that list silently widens who and what can read every document inside, retroactively, with no event that anyone would notice. So when a flow proposes a change to that list, the agent stops and puts the specifics in front of the collection owner: "Collection HR Policies currently answers to two models. Adding the requested model ID makes all 340 documents in it readable by that model, including the 12 marked personnel. Approve, approve with the personnel items removed to a separate collection, or reject?" The owner decides. The same gate covers Delete Document Collection, which removes the database and every document in it permanently, and where the honest recommendation is usually the one nele.ai itself makes: disable the collection instead, and keep the option to change your mind. Agents handle the ingestion, the embedding and the search. A person owns the permission boundary.
Agent Capabilities
20 actionsChat
1- Create Chat Completion Generates a complete response from any model enabled in the workspace. Optionally grounds the answer in one of your document collections for retrieval-augmented answers, or enables web search for current information. The two are mutually exclusive, so the flow chooses which mode a given question needs. Streaming is not exposed, so the whole answer returns in one response.
Images and Audio
3- Generate Image Generates an image from a text prompt using DALL-E, GPT Image, Google Imagen or Gemini models hosted through nele.ai. Returns a temporary URL with the image dimensions and credit usage. Quality, size, style, background and aspect ratio support differs per model, and unsupported options are ignored by the provider.
- Upload Image Attachment Uploads a PNG, JPG or WebP image fetched from a URL and returns an attachment ID that can be referenced from a later chat message. Used when a person needs to show the agent something rather than describe it.
- Transcribe Audio Transcribes an audio file into text using Azure Whisper or Azure Fast Transcription, fetching the file from a URL. Accepts aac, flac, mp3, mpeg, mpga, m4a, ogg, wav and webm up to 500 MB. Supplying the source language improves accuracy.
Document Collections
13- List Document Collections Returns a paginated list of the knowledge databases available to the account, with free-text search and sorting by title or description. Used to resolve the right collection before searching or grounding a completion.
- Create Document Collection Creates a new knowledge database assigned to the current user, with an explicit list of model IDs permitted to query it. Used to stand up a scoped collection per department, project or sensitivity level.
- Get Document Collection Retrieves one knowledge database with its title, description, enabled state and permitted models. Used to read the current permission boundary before a flow proposes changing it.
- Update Document Collection Updates the title, description, enabled state or permitted models of an existing knowledge database. Disabling a collection keeps its documents but stops it being used for retrieval, which makes it the reversible alternative to deletion.
- Delete Document Collection Permanently deletes a knowledge database and every document inside it. This cannot be undone, which is why flows propose it rather than run it.
- Search Document Collection Runs a semantic vector search against one knowledge database and returns matching snippets with their source filenames and any context data stored with them. Raising the maximum distance returns more results with less precision.
- Upload Document To Collection Downloads a document from a URL and adds it to a knowledge database. Auto embed indexes it immediately; without it, run Embed Document Collection afterwards. Context strings supplied here come back alongside future search results.
- Add Document From URL Has nele.ai fetch the document directly from a URL, optionally using basic, bearer or custom-header authentication to reach a protected source. Used when the file lives behind an internal gateway and should not be routed through the flow.
- List Document Collection Items Returns a paginated list of the documents in a knowledge database. Used to review what has been indexed and to pick item IDs for the get, embed or delete actions.
- Get Document Collection Item Retrieves one stored document with its filename, embedding state and any context strings recorded when it was added. Used to confirm a specific file finished indexing.
- Delete Document Collection Item Permanently removes one document and its embeddings from a knowledge database. The rest of the collection is unaffected.
- Embed Document Collection Generates embeddings for every not-yet-indexed document in a collection, making them searchable. Run after uploading without auto embed. Large collections take time to finish.
- Embed Document Collection Item Generates embeddings for one specific document without reprocessing the rest of the collection. Used after replacing or re-uploading a single file.
Account
3- List Models Lists every model available to the current user, split into personal and team models, along with the document collections usable with them. Each entry reports availability, the serving company and provider, and credit consumption.
- Get User Settings Retrieves the settings of the user the API key belongs to, including which modules are enabled. Used as a pre-flight check that features such as document collections are available before calling them.
- List User Prompts Lists the saved prompts belonging to the API key's user. Used to reuse a prompt an operator already tuned in the product rather than duplicating it in the flow.
Frequently Asked Questions
What can FlowRunner do with nele.ai?
FlowRunner agents can run Create Chat Completion, Generate Image, and Transcribe Audio in nele.ai, plus 17 more actions.
Does connecting nele.ai to FlowRunner require OAuth?
No. nele.ai connects to FlowRunner with an API key, no OAuth flow required.
Can nele.ai trigger a FlowRunner workflow automatically?
nele.ai doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with nele.ai
$100 in credits. No card required. Connect in minutes.