Tokenomics & Resource Capping
Understanding token quotas, parent-child pooling, model weightings, and resource limitations.
To ensure high availability and fair allocation of GPU and reasoning resources, AutomateIt implements a cost-weighted token tracking and resource capping system across all subscription tiers.
1. Token Quota & Allocation
Each plan tier is allocated a monthly token quota to cover Large Language Model (LLM) reasoning and vector search operations. If you are using our Bundled AI model (default), your workspace consumes tokens from this pool.
| Plan Tier | Included Monthly Tokens | Included AI Conversations | Target Operations |
|---|---|---|---|
| Starter | 3,000,000 | ~1,000 conversations | Low-volume micro-businesses |
| Pro | 10,000,000 | Unlimited | Medium-volume teams |
| Agency | 100,000,000 | Unlimited | Multi-tenant agency managers |
| Enterprise | 1,000,000,000 | Unlimited | Custom high-volume organizations |
2. Parent-Child Workspace Pooling
For digital agencies and multi-tenant setups, token quotas are pooled globally across the parent account and all sibling/child workspaces:
- Shared Token Reservoir: Sibling workspaces do not need individual billing keys. They consume tokens directly from the parent tenant's centralized monthly pool.
- Centralized Control: The parent tenant can monitor real-time token utilization across all active client workspaces from the Agency Billing panel.
3. Cost-Weighted Token Tracking
Different Large Language Models have varying execution costs. To reflect this, AutomateIt tracks consumption using model-specific cost multipliers applied to raw input/output tokens:
$$\text{Tokens Charged} = (\text{Input Tokens} \times \text{Input Multiplier}) + (\text{Output Tokens} \times \text{Output Multiplier})$$
Multiplier Reference Table
- Standard Models (e.g., Llama 3.1 70B, Gemma 3 27B):
- Input: $1.0\times$
- Output: $1.0\times$
- Premium Models (e.g., Claude 3.5 Sonnet, GPT-4o):
- Input: $3.0\times$
- Output: $3.0\times$
- Advanced Reasoning Models (e.g., o1, Gemini 1.5 Pro):
- Input: $5.0\times$
- Output: $5.0\times$
4. BYOK (Bring Your Own Key) Bypass
High-volume workspaces can connect their own developer API keys (OpenAI, Anthropic, or custom endpoint) in Settings → AI Provider:
- Bypass Quota Limits: Once a custom key (BYOK) is enabled, your workspace is exempted from daily/monthly token quota checks.
- Direct Wholesale Billing: You pay the LLM providers directly for raw token usage, saving markup costs on high volume.
5. Plan Feature & Resource Caps
Features and active integrations are capped according to your plan tier. These limits are checked on the frontend and enforced strictly on the backend:
| Resource / Feature | Starter | Pro | Agency | Enterprise |
|---|---|---|---|---|
| Team Seats | 1 seat | 5 seats | Unlimited | Unlimited |
| Connected WABA Numbers | 1 number | 3 numbers | 20 numbers | Unlimited |
| Active custom API Tools | 2 tools | 6 tools | Unlimited | Unlimited |
| Specialist Playbook Skills | 5 skills | 20 skills | Unlimited | Unlimited |
| Knowledge Base Sources | 5 sources | 20 sources | Unlimited | Unlimited |
| Broadcast Target Limit | 200 contacts/run | Unlimited | Unlimited | Unlimited |
| Long-Term Memory Vault | ❌ Disabled | ✅ Enabled | ✅ Enabled | ✅ Enabled |
How Resource Exceedance is Handled
- Frontend Gated Control States: Input fields, upload buttons, and checkboxes are dynamically disabled with lock icons and upgrade tooltips once your plan limit is reached.
- RAG Slicing: If a workspace has more than the allowed number of active knowledge base files (e.g. upload files), the system slices the search index to query only the first $N$ allowed sources (e.g., 5 for Starter).
- Soft Warnings: Real-time email and in-app alerts are dispatched when any resource utilization reaches 80%.