build+docs: Make ollama optional (#671)

docker-compose.ollama.yml

@@ -0,0 +1,10 @@
+services:
+  ollama:
+    image: ollama/ollama:${OLLAMA__VERSION}
+    ports:
+      - 11434:11434
+    volumes:
+      - ollama:/root/.ollama
+
+volumes:
+  ollama:

docker-compose.yml

@@ -172,21 +172,9 @@ services:
     depends_on:
       - temporal_postgres_db
 
-  ollama:
-    container_name: ollama
-    tty: true
-    restart: unless-stopped
-    networks:
-      - core
-    image: ollama/ollama:${OLLAMA__VERSION}
-    ports:
-      - 11434:11434
-    volumes:
-      - ollama:/root/.ollama
 volumes:
   core-db:
   temporal-db:
-  ollama:
 
 networks:
   core:

docs/mint.json

@@ -90,7 +90,8 @@
           "group": "Deployment Options",
           "pages": [
             "self-hosting/deployment-options/docker-compose",
-            "self-hosting/deployment-options/aws-ecs"
+            "self-hosting/deployment-options/aws-ecs",
+            "self-hosting/deployment-options/ollama"
           ]
         },
         {

docs/self-hosting/deployment-options/ollama.mdx

@@ -0,0 +1,47 @@
+---
+title: Self-hosted LLMs
+description: Learn how to self-host LLMs in Tracecat with Ollama.
+---
+
+## Prerequisites
+
+- Basic Tracecat deployment with [Docker Compose](/self-hosting/deployment-options/docker-compose)
+- Minimum 10GB of disk space
+
+## Instructions
+
+<Note>
+    Deploying self-hosted LLMs is resource intensive with large downloads and large model weights.
+    The Ollama Docker image is 1.5GB+ large and model weights can vary greatly in size.
+
+    Only models less than 5GB in size are currently supported.
+</Note>
+
+Tracecat supports self-hosted LLMs through [Ollama](https://ollama.ai/).
+
+Supported models:
+
+- [`llama3.2`](https://ollama.com/library/llama3.2): 1.3GB
+- [`llama3.2:1b`](https://ollama.com/library/llama3.2): 2.0GB
+
+<Steps>
+    <Step title="Configure open source models">
+        Specify the open source models you wish to use in Tracecat by setting the `TRACECAT__PRELOAD_OSS_MODELS` environment variable in the `.env` file.
+
+        For example, to preload the `llama3.2` model, set the following:
+        ```
+        TRACECAT__PRELOAD_OSS_MODELS=llama3.2
+        ```
+    </Step>
+    <Step title="Deploy">
+        Deploy Tracecat with the Ollama docker compose extension:
+        ```bash
+        docker compose up -f docker-compose.yml -f docker-compose.ollama.yml up -d
+        ```
+    </Step>
+    <Step title="AI Action">
+        You can now use Tracecat's AI action to call your preloaded open source LLMs.
+        For example, to call the `llama3.2` model, you can specify the following arguments:
+        ![AI Action](/img/self-hosting/ai-action.png)
+    </Step>
+</Steps>

docs/self-hosting/introduction.mdx

@@ -14,3 +14,6 @@ Choose from a number of deployment options listed below to get started.
     Use Terraform to deploy Tracecat into ECS Fargate.
   </Card>
 </CardGroup>
+
+Interested in using open source LLMs (e.g. llama3.1) in Tracecat's AI actions?
+Check out our guide on deploying [self-hosted LLMs](/self-hosting/deployment-options/ollama) with Tracecat's Ollama Docker Compose extension.