Ollama

LogicGrid talks to a local Ollama server over its REST API. By default the client connects to http://localhost:11434.

• Install: ollama.ai
• Models library: ollama.com/library

Pull a model once:

ollama pull llama3.2

Use it from LogicGrid

There are two equivalent ways to instantiate the Ollama LLM client.

Option 1 — static factory (recommended)

using LogicGrid.Core.Llm;

var llm = LlmClientBase.Ollama("llama3.2");

Custom base URL (Ollama on another host or in Docker):

var llm = LlmClientBase.Ollama(
    model:   "llama3.2",
    baseUrl: "http://192.168.1.50:11434");

Parameter	Type	Default	Notes
model	string	"llama3.2"	Ollama model tag. Pull it first with ollama pull <model>.
baseUrl	string	"http://localhost:11434"	Ollama HTTP endpoint. Override for Docker, remote hosts, or reverse proxies.

Option 2 — direct construction

using LogicGrid.Core.Providers;

var llm = new OllamaClient(
    baseUrl:      "http://localhost:11434",
    defaultModel: "llama3.2");

Parameter	Type	Default	Notes
baseUrl	string	"http://localhost:11434"	Same as the factory's baseUrl.
defaultModel	string	"llama3.2"	The model used when the agent or call site doesn't override it.

The factory and the constructor produce equivalent clients — the factory is just a more discoverable entry point. Use direct construction when you need a custom HttpClient (next section) or are wiring up DI by hand.

Custom HttpClient (proxies, timeouts, headers)

OllamaClient accepts a pre-configured HttpClient. Use this when you need a corporate proxy, a longer timeout, or a custom header (for example, when fronting Ollama with a reverse proxy that requires authentication).

using System.Net;
using System.Net.Http;
using LogicGrid.Core.Providers;

var handler = new HttpClientHandler
{
    Proxy    = new WebProxy("http://corporate-proxy:8080"),
    UseProxy = true,
};

var http = new HttpClient(handler)
{
    Timeout = TimeSpan.FromMinutes(5),
};
http.DefaultRequestHeaders.Add("X-Forwarded-User", "service-account");

var llm = new OllamaClient(
    httpClient:   http,
    baseUrl:      "https://ollama.internal.example",
    defaultModel: "llama3.2");

Tool calling

LogicGrid ships two tool-calling strategies:

• PromptSchemaStrategy (default) — describes tools in the system prompt and parses JSON tool calls out of the LLM's text response. Works with every model.
• NativeToolCallingStrategy — uses the provider's native tool-call API. Stronger when the model supports it.

Some Ollama models support native tool calling (qwen2.5, llama3.1, llama3.2 and others); many do not. Stay on the prompt-schema default unless you've verified the specific model.

Full example — calculator tool over Ollama

using LogicGrid.Core.Agents;
using LogicGrid.Core.Llm;
using LogicGrid.Core.Tools;
using LogicGrid.Tools.Tools;

var llm = LlmClientBase.Ollama("llama3.2");

IAgent math = new Agent<string>(
    name:         "Mathlete",
    description:  "Solves arithmetic.",
    systemPrompt: "Use the calculator tool when the user asks for a number.",
    llm:          llm,
    tools:        new ToolBase[] { new CalculatorTool() });

Console.WriteLine(
    await math.RunAsync(
        input: "What is (17 * 23) + 91?",
        ctx:   new AgentContext().WithLogging()));

09:14:02.118 [INF] [Mathlete] started | input: What is (17 * 23) + 91?
09:14:02.420 [INF] [Mathlete] tool call → calculator { "expression": "(17 * 23) + 91" }
09:14:02.430 [INF] [Mathlete] tool result | 482
09:14:03.180 [INF] [Mathlete] completed | output: (17 * 23) + 91 = 482.

(17 * 23) + 91 = 482.

For the full strategy reference and how to switch to native, see Tool calling strategy.

Embeddings

Same two-way pattern as the LLM client.

Option 1 — static factory (recommended)

using LogicGrid.Memory.Embeddings;

var embedder = EmbeddingClientBase.Ollama("nomic-embed-text");
var vector   = await embedder.EmbedAsync("Hello, world.");

Parameter	Type	Default	Notes
model	string	"nomic-embed-text"	Embedding model tag. Recommended: nomic-embed-text (768 dims), mxbai-embed-large (1024 dims), bge-m3 (multilingual).
baseUrl	string	"http://localhost:11434"	Ollama HTTP endpoint.
dimensions	int	0	Vector size. 0 = discover on first call (one extra round-trip). Pass the known dimension to skip discovery.

Option 2 — direct construction

Mirrors OllamaClient exactly: baseUrl first, then defaultModel.

using LogicGrid.Memory.Embeddings;

var embedder = new OllamaEmbeddingClient(
    baseUrl:      "http://localhost:11434",
    defaultModel: "nomic-embed-text",
    dimensions:   768);

Parameter	Type	Default	Notes
baseUrl	string	"http://localhost:11434"	Ollama HTTP endpoint.
defaultModel	string	"nomic-embed-text"	Embedding model used when the call site doesn't override it.
dimensions	int	0	Vector size. 0 = discover on first call.

A second overload accepts a custom HttpClient for proxies, longer timeouts, or DI:

using System.Net.Http;
using LogicGrid.Memory.Embeddings;

var http = new HttpClient { Timeout = TimeSpan.FromMinutes(2) };

var embedder = new OllamaEmbeddingClient(
    httpClient:   http,
    baseUrl:      "http://localhost:11434",
    defaultModel: "nomic-embed-text",
    dimensions:   768);

Parameter	Type	Default	Notes
httpClient	HttpClient	(required)	Pre-configured client.
baseUrl	string	"http://localhost:11434"	Same as above.
defaultModel	string	"nomic-embed-text"	Same as above.
dimensions	int	0	Same as above.

Troubleshooting

• Connection refused — Ollama isn't running. Start it from the menu bar or ollama serve.
• model not found — Run ollama pull <model> first.
• Slow first request — Models are loaded lazily on first call. Run a warm-up CallAsync at startup if first-request latency matters.