Configuring Clara AI

Use Launch AI Server Locally to run the Clara AI Agent on the Striim host, or use Use Existing Server to connect to an agent that you deploy on another host. The following sections describe these options.

The Launch button remains disabled until all required fields are complete. Striim starts the Clara AI Agent containers and connects the Ask Clara panel to the configured backend.

The remote host must have:

Copy the supplied Compose file to the remote host:

scp striim-ai/docker-compose.yml <user>@<remote-host>:~/striim-ai/

On the remote host, run:

docker login artifacts.striim.com --username <cluster-name>

Enter the Striim product key at the password prompt. Do not place the product key directly in a command or script.

Configure the following environment variables:

The embedding model is required when ENABLE_KNOWLEDGEBASE=true. If you omit an embedding-provider setting, Clara uses the corresponding chat-provider setting. EMBEDDING_MODEL does not have a fallback and must be set explicitly.

The following example configures the agent to use OpenAI. Replace each placeholder and use a model supported by your provider.

export AI_PROVIDER=openai
export AI_API_KEY=<provider-api-key>
export AI_MODEL=gpt-5.4-nano
export EMBEDDING_MODEL=text-embedding-3-small
export COPILOT_TOKEN=<random-uuid>
export STRIIM_BASE_URL=http://<striim-platform-host>:9080

docker compose -f ~/striim-ai/docker-compose.yml up -d

Use your organization's secret-management mechanism to supply provider keys and tokens in production environments.

Run the following commands on the Clara AI Agent host:

docker compose -f ~/striim-ai/docker-compose.yml ps
docker compose -f ~/striim-ai/docker-compose.yml logs striim-ai-agent
curl http://localhost:8001/health

The health endpoint is available at http://localhost:8001/health. The Clara UI is available at http://<remote-host>:9090.

A green confirmation banner indicates that Clara is connected and ready. You can update the configuration at any time by returning to this panel.

The supported configuration uses qwen3.6:35b for chat and completion and nomic-embed-text for embeddings. This configuration supports documentation search and basic cluster operations.

Model quality, response time, and reliability depend on the available GPU memory, context length, concurrency, and workload.

Provision an Ollama host that meets the following requirements:

The supported chat model requires approximately 23 GB of storage and approximately 25 GB of GPU memory when loaded. Additional GPU memory is required for context and concurrent requests.

The following cloud instances provide starting configurations for an Ollama host:

Instance availability, quotas, and pricing vary by region and account. Confirm current capacity with the cloud provider before provisioning.

Ollama allocates additional GPU memory as concurrency and context length increase:

Required GPU memory increases approximately with the number of parallel requests multiplied by the configured context length. Provision capacity for the expected number of concurrent Clara users and load-test the configuration while monitoring GPU memory with nvidia-smi.

Connect to the Ollama host and verify that the GPU driver is active:

nvidia-smi

Install Ollama:

curl -fsSL https://ollama.com/install.sh | sh

Create a systemd override:

sudo mkdir -p /etc/systemd/system/ollama.service.d
sudo tee /etc/systemd/system/ollama.service.d/override.conf > /dev/null <<'EOF'
[Service]
Environment="OLLAMA_HOST=0.0.0.0"
Environment="OLLAMA_NUM_PARALLEL=4"
Environment="OLLAMA_CONTEXT_LENGTH=8192"
Environment="OLLAMA_KEEP_ALIVE=-1"
EOF

sudo systemctl daemon-reload
sudo systemctl restart ollama
sudo systemctl enable ollama

The configuration uses the following settings:

ollama pull qwen3.6:35b
ollama pull nomic-embed-text

Test chat generation:

curl http://localhost:11434/api/generate -d '{
  "model": "qwen3.6:35b",
  "prompt": "Say hello world",
  "stream": false,
  "think": false
}'

Test embeddings:

curl http://localhost:11434/api/embeddings -d '{
  "model": "nomic-embed-text",
  "prompt": "hello"
}'

Confirm model and GPU status:

nvidia-smi --query-gpu=memory.used,memory.total --format=csv
ollama ps

From the Clara AI Agent host, verify that the Ollama endpoint is reachable:

curl http://<ollama-private-ip>:11434/api/tags

An HTTP 200 response that lists both models confirms that the network rule and OLLAMA_HOST setting allow the Clara AI Agent to reach Ollama.

Use the following values when you configure the Clara AI Agent:

Restart the Clara AI Agent after changing its provider configuration and then verify the connection from Ask Clara.

Symptom: Clara reports that the AI provider configuration is invalid or the provider rejects the credentials.

Resolution:

Symptom: Clara reports that the chat model is unavailable, and the agent log records a 404 NOT_FOUND response from the provider.

Resolution: The provider has renamed or retired the model. Check the provider's current model list, then enter the current identifier in AI Model or in AI_MODEL. For Google Gemini, see Gemini API / Models.

Symptom: Clara is configured, but the Ask Clara panel cannot reach the agent.

Resolution:

Symptom: The supported chat model does not load or the Ollama process stops under load.

Resolution:

Symptom: A request to the Ollama endpoint returns a connection error.

Resolution:

Symptom: Clara can respond, but documentation-search results are missing or incomplete shortly after setup.

Resolution: Allow up to 10 minutes for the initial knowledgebase index to finish building. If the issue continues, verify that ENABLE_KNOWLEDGEBASE=true and that the embedding model is configured.