Ollama

Run open-source models like Llama, Qwen, DeepSeek, and Gemma on your own hardware and call them directly from flows. Generation, embeddings, and model management with no data leaving your infrastructure.

10 actions available

Sensitive internal document queued for classification

Agent sends the document to a local Ollama model

Agent reads the classification and extracted fields

Agent routes by classification without the data leaving the network

Agent files the document with its classification

The owning team gets a summary of the batch

A document the model is unsure about routes to a person to label

What This Integration Enables

Ollama runs open models locally and exposes them to a flow: generation for text tasks, embeddings for retrieval, and model management to pull and organize the models you run. Because the model runs on your own hardware, the data being processed never leaves your network, and there is no per-call inference bill. For regulated or air-gapped environments, local inference is not a preference but a requirement. The value is being able to build the same kind of AI workflow you would with a cloud model, entirely on-prem. An orchestration layer runs that workflow and inserts a person where judgment is needed, and FlowRunner is built for that layer, including self-hosted deployment so the whole flow stays inside your network.

Without FlowRunner

Data leaves the network Sensitive text sent to an external model provider

Per-call inference cost Every model call metered by a vendor

No local model option On-prem requirements rule out cloud-only AI

With FlowRunner

Inference on your hardware Models run where the data already lives

No per-call fees Inference cost is your own compute, not a vendor bill

On-prem AI Open models run inside your own network

Use Case Scenarios

On-Prem Document Classification

Sensitive documents cannot leave the network but still need classifying. The agent sends each to a local Ollama model, reads the classification, and routes accordingly, all inside your own infrastructure. Confident classifications flow through; anything the model is unsure about routes to a person. The data never touches an external provider.

Private Knowledge Retrieval

An internal knowledge base holds information that cannot go to a cloud model. The agent uses local Ollama embeddings to index it and retrieve answers, keeping the entire retrieval pipeline on-prem. Employees get grounded answers, and questions touching the most sensitive material still route to the owning team.

Cost-Controlled Bulk Processing

A high volume of routine text tasks would run up a large per-call bill on a cloud provider. The agent runs them against a local Ollama model instead, trading vendor fees for your own compute. The work gets done at a predictable cost, and exceptions the model flags are routed to a person.

Human-in-Loop Highlight

Running a model on your own hardware changes where the data goes, not whether its output needs judgment. When a local Ollama model returns a low-confidence classification, or produces output that drives a real decision, FlowRunner routes it through a [human-in-the-loop](/concepts/human-in-the-loop/) step: the agent pauses, presents the output and its context, and sends it to the right person via Slack. They decide. Local inference keeps the data in your network; a person still owns the calls that matter, and with self-hosted FlowRunner the entire flow stays on-prem.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via Slack

Human decides

Agent resumes with decision

Agent Capabilities

10 actions

Generation

Generate Completion Generates a single completion for a prompt using a local Ollama model. Supports JSON mode and structured outputs (JSON schema), thinking control for reasoning models (e.g. deepseek-r1, qwen3), and raw model options like temperature and context size.
Chat Sends a multi-turn conversation to a local Ollama model and returns the next assistant message.

Embeddings

Create Embeddings Generates embedding vectors for one or more input texts using a local embedding model (e.g. 'nomic-embed-text', 'mxbai-embed-large', 'all-minilm').

Model Management

List Local Models Lists all models available locally on the Ollama server, including name, size in bytes, digest, last modified date and details such as family, parameter size and quantization level.
Show Model Info Retrieves detailed information about a local model: its Modelfile, default parameters, prompt template, details (family, parameter size, quantization) and capabilities (e.g. 'completion', 'vision', 'tools', 'thinking').
Pull Model Downloads a model from the Ollama library (https://ollama. com/library) to the server, e.g. 'llama3.
Copy Model Copies an existing local model to a new name, e.g. to tag a custom variant before editing.
Delete Model Permanently deletes a model and its data from the Ollama server's local storage. Fails if the model does not exist.
List Running Models Lists the models currently loaded into memory on the Ollama server, including total size, VRAM usage (size_vram) and when each model is scheduled to be unloaded (expires_at).

Server

Get Version Retrieves the version of the Ollama server. Also useful as a connectivity check for the configured server URL.

Start building with Ollama

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us