Models

Models define the LLM configurations that power AI student personas during office hour simulations. Each model wraps a specific LLM (such as gpt-4o or claude-3-opus) with settings for temperature, reasoning level, modalities, and pricing, and links back to the provider that supplies the underlying API.

What is a Model?

A model in Glow represents a configured LLM identity. It connects an external language model to the platform by pairing it with a provider, quality tier, and feature flags. Models are then attached to agents through the agent_models_junction, giving each agent a specific conversational capability during simulations.

For example, a university might configure a gpt-4o-sim model with medium temperature and text-only modality for TA training conversations, and a separate claude-3-haiku-quick model for rapid policy-lookup agents.

How it Connects

Provider  -->  Model  -->  Agent (via agent_models_junction)
                |
                +--> Eval (via eval model_ids)

• Provider supplies the LLM backend (API keys, endpoints). Every model has a provider_id.
• Agent uses one or more models through the agent_models_junction table. The model determines which LLM powers that agent’s conversations.
• Eval references models through model_ids so that evaluation runs can score performance across different LLM configurations.

Create a model

Via the CLI

Calls below use $GLOW_INSTANCE_URL + $GLOW_TOKEN — see Authentication to export them once.

glow models create --body '{
  "models": [{
    "name": "gpt-4o-sim",
    "description": "GPT-4o configured for student persona simulations",
    "provider_ids": ["PROVIDER_UUID"]
  }]
}'

Via the API

curl -X POST $GLOW_INSTANCE_URL/model/create \
  -H "Authorization: Bearer $GLOW_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "models": [{
      "name": "gpt-4o-sim",
      "description": "GPT-4o for TA training simulations",
      "provider_ids": ["PROVIDER_UUID"]
    }]
  }'

Each entry returns the new model_id, the generated draft_id, and the initial version.

Provider linking

Every model requires at least one provider. The provider_ids field connects the model to the provider that holds the API endpoint and credentials. When you search models, the response includes provider_id and provider_name for each entry, along with filter_provider_ids for filtering.

# Filter models by provider
curl -X POST $GLOW_INSTANCE_URL/model/search \
  -H "Authorization: Bearer $GLOW_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"filter_provider_ids": ["OPENAI_PROVIDER_UUID"]}'

Capabilities and feature flags

Models support several capability dimensions configured through resource IDs:

Use flag_ids to toggle boolean capabilities like image_model or active/inactive status.

The draft cycle

Models support a draft workflow for staging changes before publishing. Use the draft system to preview configuration changes without affecting running simulations.

# Create or update a draft
glow models draft --body '{
  "name": "gpt-4o-sim-v2",
  "description": "Updated temperature for more natural conversation",
  "temperature_level_ids": ["MEDIUM_TEMP_UUID"],
  "provider_ids": ["OPENAI_PROVIDER_UUID"]
}'
 
# List your drafts
glow models list

The draft endpoint uses PATCH semantics with optimistic concurrency via expected_version. The response returns a draft_id, new_version, and the server-authoritative form_state.

Search & filter

# Free-text search
glow models search --body '{"search": "gpt-4o"}'
 
# Facet filter: only OpenAI text models in CS
glow models search --body '{
  "filter_provider_ids": ["OPENAI_PROVIDER_UUID"],
  "filter_department_ids": ["CS_DEPT_UUID"],
  "page_size": 25
}'

The response includes a department_filter, provider_filter, and modality facets with available options, counts, and selected IDs for building filter UIs.

Bulk operations

Bulk delete and update follow the canonical all-matching shape — explicit IDs, or all: true with flat filter fields plus optional excluded_ids and a patch body. The persona surface is the prove-out; models follow the same shape.

Delete by explicit IDs:

glow models delete --body '{
  "model_ids": ["model-1", "model-2"]
}'

Delete all matching a filter (with exclusions):

glow models delete --body '{
  "all": true,
  "filter_provider_ids": ["deprecated-provider-uuid"],
  "excluded_ids": ["model-still-in-use"]
}'

Bulk update via patch:

glow models update --body '{
  "all": true,
  "filter_provider_ids": ["old-provider-uuid"],
  "patch": { "provider_ids": ["new-provider-uuid"] }
}'

Common Operations

Related

• Models API Reference — full endpoint and type documentation
• Models CLI Reference — all CLI commands
• Providers Guide — set up the LLM backend a model connects to
• Evals Guide — evaluate model performance across simulations