Credits
Credits make AI Grid Agent and chat runs predictable before you launch them.
What Are Credits?​
Credits are the customer-facing balance used when Lampi runs AI work for your organization. The estimate is shown before execution and is based on the request type, selected skills/tools, selected model, and number of cells or messages being run.
Key principles:
- Credits are estimated before execution
- Pricing uses fixed tiers instead of unpredictable tool-call counts
- Higher-cost LLMs and higher reasoning effort can apply configured multipliers
- Reserved credits are refunded when a task is stopped, canceled, or fails because of a system issue
- Provider usage cost is informational only and is separate from your fixed credit balance
Fixed Credit Tiers​
| Operation Tier | Base Credits |
|---|---|
| Basic LLM | 1 credit |
| Fast Retrieval | 2 credits |
| Standard Skill or Tool | 5 credits |
| Slide Deck | 15 credits |
| DeepSearch | 100 credits |
The most expensive matching tier is used for each question or chat request. For example, a question that combines a slide skill and web retrieval is priced as a Slide Deck request, not as two separate charges.
Model And Reasoning Multipliers​
Admins configure model multipliers in LLM Settings.
| Model Family | Default Multiplier |
|---|---|
| All models | x1.0 |
| GPT 5.4 | x1.5 |
| GPT 5.5 | x2.0 |
Models that support reasoning can also apply a reasoning-effort multiplier. The selector shows the combined multiplier next to the selected model.
| Reasoning Effort | Multiplier |
|---|---|
| Off | x1.0 |
| Low | x1.0 |
| Medium | x1.0 |
| High | x1.5 |
| Extra high | x2.0 |
The final unit price is rounded up:
credits = ceil(base tier credits × model multiplier × reasoning multiplier)
Examples:
- Basic LLM on GPT 5.4:
ceil(1 × 1.5)= 2 credits - Standard Skill on GPT 5.4:
ceil(5 × 1.5)= 8 credits - Standard Skill on GPT 5.4 with High reasoning:
ceil(5 × 1.5 × 1.5)= 12 credits - Standard Skill on GPT 5.4 with Extra high reasoning:
ceil(5 × 1.5 × 2.0)= 15 credits - DeepSearch on GPT 5.5:
ceil(100 × 2.0)= 200 credits
DeepSearch already includes high-reasoning search behavior in its fixed tier. Reasoning effort does not add an extra reasoning multiplier to DeepSearch.
How Credit Calculation Works​
| Operation Type | Credit Calculation |
|---|---|
| Single Chat | One unit at the detected tier, multiplied by the selected model and reasoning multipliers |
| Single Cell | One unit for that cell's question tier, model multiplier, and reasoning multiplier |
| Column Re-run | Unit price multiplied by the number of rows in that column |
| Full Agent Run | Sum of all selected generated questions × selected rows, skipping existing answers unless restarting |
Credits are reserved before the run begins. When the run succeeds, the reservation is finalized. When the run is stopped, canceled, or fails before completion, the reserved credits are returned.
Credit Management​
| Feature | Description |
|---|---|
| Organization Balance | Credits are shared across your organization |
| Pre-Run Estimates | The UI shows the estimated cost before large Agent Grid runs |
| LLM Settings Multipliers | Super admins can configure model-specific multipliers; supported reasoning effort adds the published reasoning multiplier |
| Free Plan Credits | Free credits do not renew monthly. Once they reach 0, paid actions stay blocked until an admin tops up or upgrades |
| Paid Plan Credits | Paid plans grant a monthly credit allowance that resets on the Stripe billing cycle |
| Credit Top-Ups | Top-ups add a fixed 1,000 credits without changing the subscription plan |
The usage-cost estimate shown in billing is provider telemetry for transparency. It is not converted into credits and does not change the fixed credits charged for an action.
Credit Expiration Rules​
| Account State | What Happens |
|---|---|
| Free plan | Credits are not renewed monthly. Top-up credits remain available until consumed |
| Paid plan | Monthly credits reset on each paid billing cycle. Remaining top-up credits are part of the current cycle balance and should be used before that reset |
| Canceled, deleted, or refunded paid subscription | The workspace is downgraded to Free and remaining credits are reset to 0 |
When Credits Are Refunded​
Credits are returned for:
- Stopped Tasks - User stops a running task
- Canceled Tasks - User cancels before completion
- System Failures - Technical failures interrupt execution
- Quota Reservation Failures - The run does not start because credits could not be reserved
Credit Usage Tips​
- Start Small - Test with a few rows before running full datasets
- Review Estimates - Check the estimate before launching large grid operations
- Use Targeted Runs - Run specific rows, columns, or cells when updating data
- Choose Models Intentionally - GPT 5.4, GPT 5.5, and higher reasoning effort consume more credits through multipliers
- Avoid Unneeded Skills - Skills move requests into higher tiers when they add extra capabilities
Managing Your Credit Balance​
If your organization runs out of credits, workspace admins can buy a fixed 1,000-credit top-up from Usage & Billing or upgrade to a paid monthly plan. Non-admin users should contact a workspace admin.
Next: Slides →