Technical Alignment and Proposal Outline features— are powered by an LLM provider configured at deploy time. Configuration is workspace-agnostic: it applies to the whole VisibleThread deployment, not per-workspace.

Three providers are supported: Amazon Bedrock, Azure OpenAI and LiteLLM. If the relevant properties are left unset, Technical Analysis and Outline Generation features are treated as not configured and remain disabled — there are no defaults that silently enable them.

Where the properties go

Regardless of which provider you use, the properties are set as environment variables in the VisibleThread environment file:

After editing the file, restart the relevant VisibleThread service(s) for the changes to take effect.

Option 1: Amazon Bedrock

Add the following example to visiblethread.env:

vt.llm.analysis.provider=${VT_LLM_ANALYSIS_PROVIDER:bedrock}
vt.llm.analysis.bedrock.region=${VT_LLM_ANALYSIS_BEDROCK_REGION:us-west-2}
vt.llm.analysis.bedrock.model=${VT_LLM_ANALYSIS_BEDROCK_MODEL:global.anthropic.claude-sonnet-4-6}
vt.llm.analysis.bedrock.timeout-seconds=${VT_LLM_ANALYSIS_BEDROCK_TIMEOUT_SECONDS:300}

Property reference:

• vt.llm.analysis.provider — must be bedrock to select this provider.
• vt.llm.analysis.bedrock.region — the AWS region hosting the Bedrock endpoint (e.g. us-west-2).
• vt.llm.analysis.bedrock.model — the Bedrock model identifier to invoke.
• vt.llm.analysis.bedrock.timeout-seconds — request timeout, in seconds, for calls to Bedrock.

Notes:

• The instance/role running VisibleThread must have IAM permissions to invoke the specified model in Bedrock in the configured region.
• No API key is set here — Bedrock auth relies on the AWS credentials/role available to the host (instance profile, SSM, etc.), consistent with how other AWS-integrated VT services authenticate.

Option 2: Azure OpenAI

Add the following example to visiblethread.env:

vt.llm.analysis.provider=${VT_LLM_ANALYSIS_PROVIDER:azure-openai}
vt.llm.analysis.azure-openai.endpoint=${VT_LLM_ANALYSIS_AZURE_OPENAI_ENDPOINT:https://example.cognitiveservices.azure.com}
vt.llm.analysis.azure-openai.api-key=${VT_LLM_ANALYSIS_AZURE_OPENAI_API_KEY:xxxxxxxxxxxxxxxxxxxxxxxxxxxxx}
vt.llm.analysis.azure-openai.model=${VT_LLM_ANALYSIS_AZURE_OPENAI_MODEL:gpt-5.4}

Property reference:

• vt.llm.analysis.provider — must be azure-openai to select this provider.
• vt.llm.analysis.azure-openai.ednpoint  — the Azure OpenAI (Cognitive Services) resource endpoint.
• vt.llm.analysis.azure-openai.api-key — the API key for the Azure OpenAI resource. 
• vt.llm.analysis.azure-openai.model — the deployed model name in the Azure OpenAI resource.

Option 3: LiteLLM

Add the following example to visiblethread.env:

vt.llm.analysis.provider=${VT_LLM_ANALYSIS_PROVIDER:litellm}
vt.llm.analysis.litellm.endpoint=${VT_LLM_ANALYSIS_LITELLM_ENDPOINT:https://litellm.example.com}
vt.llm.analysis.litellm.bearer-token=${VT_LLM_ANALYSIS_LITELLM_BEARER_TOKEN:sk-xxxxxxxxxxxxxxxxxxxx}
vt.llm.analysis.litellm.model=${VT_LLM_ANALYSIS_LITELLM_MODEL:<model-name>}

Property reference:

• vt.llm.analysis.provider: must be litellm to select this provider.
• vt.llm.analysis.litellm.endpoint: the base URL of your LiteLLM gateway. It must use https://. You can include or leave out a trailing /v1; both work.
• vt.llm.analysis.litellm.bearer-token: the LiteLLM key (virtual key or master key), sent as Authorization: Bearer <token>.
• vt.llm.analysis.litellm.model: the model name exactly as it is configured in your LiteLLM gateway (the model_name alias), not the underlying provider's model ID.

Notes:

VisibleThread talks to LiteLLM through its OpenAI-compatible API (/v1/chat/completions). ‚

All three LiteLLM properties are required. If one is missing, or the endpoint is not HTTPS, Technical Alignment and Proposal Outline are treated as not configured.

Use a model that is good at structured (JSON) output and has a large output limit. Outline generation depends on both.

The LiteLLM provider on the System Admin ‚ Generative AI page configures Visi chat only. Technical Alignment and Proposal Outline always use the vt.llm.analysis.* properties in this article, even if the Generative AI page is also set to LiteLLM

Applying the change

1. Edit the environment file for the target OS (paths above).
2. Set the four properties for your chosen provider only — don't mix properties from different providers in the same environment.
3. Save the file.
4. Restart the VisibleThread service(s) that consume this configuration so the new values are picked up.
5. Confirm Technical Alignment, Proposal Outline are available in the UI.

Disabling Technical Analysis and Outline Generation

To disable the features, remove or comment out vt.llm.analysis.provider (and the associated provider properties) from the environment file and restart the service. With no provider configured, Technical Analysis and Outline Generation features are treated as not configured and will not be exposed.