Appearance
Model Providers & Budgets
DMOps.ai supports multiple LLM providers. As an admin you configure which providers are available, which model each agent uses, and how much each agent is allowed to spend.
Supported providers
| Provider | Notes |
|---|---|
| AWS Bedrock | Recommended for production — no data leaves your AWS region. Supports Claude 3.x, Llama, Mistral |
| Anthropic (direct) | Claude models via Anthropic's API |
| OpenRouter | 200+ models from one endpoint — good for experimentation |
| OpenAI | GPT-4o, GPT-4 Turbo, and other OpenAI models |
| NVIDIA NIM | Llama and Nemotron models via NVIDIA's inference platform |
Adding a model provider
- Go to Settings → Model Providers.
- Click + Add Provider.
- Select the provider type.
- Enter the API key or configure AWS IAM role (for Bedrock).
- Click Save & Test.
WARNING
API keys are encrypted and stored securely. Never share or log them. If a key is compromised, rotate it immediately in the provider dashboard, then update it here.
Setting a default model
The default model is used by all agents that don't have a model override.
- Go to Settings → Model Providers.
- Click Set as Default on the provider row.
- In the Default Model dropdown, select the model to use for all agents by default.
Overriding the model for a specific agent
- Open the agent's detail panel (
Settings → Org Chart → [Agent]). - Go to the Model tab.
- Select a Provider and Model from the dropdowns.
- Click Save.
This agent will now use the selected model regardless of the tenant default.
Budgets
Budgets prevent agents from running up unexpected costs. You can set limits at three levels:
| Level | Where to set it | Effect |
|---|---|---|
| Per agent | Agent detail → Budget tab | Limits one agent's total spend |
| Per role | Settings → Roles → [Role] → Budget | Limits all agents with this role |
| Tenant-wide | Settings → Billing | Hard cap for the whole tenant |
Budget fields
| Field | Description |
|---|---|
| Monthly token limit | Maximum LLM tokens this agent/role can consume per calendar month |
| Monthly cost limit (USD) | Maximum spend in USD per calendar month |
| Soft warning threshold | Alert at this % of the limit (e.g. 80%) — agent continues but admin is notified |
| Hard stop | When the limit is reached, the agent stops and creates a blocked task for human review |
Viewing current spend
Go to Observability → Spending to see:
- Cost per agent, per model, and per task
- Daily spend trend
- Remaining budget for each agent
- Top tasks by cost
LLM-powered task routing
DMOps.ai uses the cheapest configured model to route new tasks to the right agent automatically. When a task is created, the dispatcher:
- Asks the LLM to score each active agent (0–100) based on the task title and description
- Applies hard capability checks (does the agent have the required integration grants and datasets?)
- Assigns the highest-scoring valid agent
This replaces keyword-matching heuristics and significantly reduces misrouting on tasks with generic or ambiguous titles.
To disable LLM routing (fallback to keyword scoring only) — go to Settings → Advanced → LLM Task Routing and toggle it off.
Cost
The routing call uses the cheapest configured model and processes only the task title and agent roster — it costs less than 1% of a typical task run.
Token costs reference
Approximate costs at time of writing (check provider pricing for current rates):
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| Claude 3.5 Haiku | $0.80 | $4.00 |
| Claude 3.7 Sonnet | $3.00 | $15.00 |
| GPT-4o mini | $0.15 | $0.60 |
| GPT-4o | $2.50 | $10.00 |
| Llama 3.3 70B (Bedrock) | $0.72 | $0.72 |
Cost optimisation
Use a cheaper model (e.g. Claude Haiku, GPT-4o mini) for high-volume routine tasks like data extraction and classification. Reserve powerful models (Sonnet, GPT-4o) for complex reasoning tasks. You can set different models per agent role to optimise spend.