DataGrout Credits Usage Examples
The Credit Calculation Model
Unit of account
One fixed number, platform wide. Enter it whichever way you think about it — all three are the same value.
$ per credit
0.002
credits per $1
500
$ per 1,000 credits
2
Charges always round up to whole credits, so this is also the smallest amount anything can cost: a call with $0.000200 of real cost still bills 1 credit ($0.002000). A smaller peg makes that rounding finer.
Platform fees
Flat charges that ride on top of cost recovery. They apply to every call that crosses the gateway, whatever the tool behind it.
Gateway base (credits)
1
Auth, materialization, policy, cadence and receipts. Never waived.
Discovery premium (credits)
2
Charged on top of the gateway base.
Discovery call (credits)
3
The base plus the premium, on a call that needs discovery.
Refraction
A new refraction is a flat 5 credits. Any model cost above the allowance is added on top of it, rounded up to a whole credit.
Refraction (credits)
5
Flat price for a new refraction.
Model allowance (credits)
0
Model calls up to this are already covered by the flat price.
| If model calls cost | Surcharge | Refraction charges |
|---|---|---|
| 0 credits | 0 | 5 |
| 1 credit | 1 | 6 |
Margin policy
Applied on top of cost recovery. Recovery itself is not tunable — an LLM call always covers its own tokens.
Profit margin (billed)
1.5
+50% on cost · 33.3% gross
Estimate bound (not billed)
1.5
Pre-flight ceiling only. Never charged.
Comped classes
1.0 means fully comped — the user pays nothing for that class of work.
Embeddings
1.0
Fully comped — the user pays nothing.
Annotations
1.0
Fully comped — the user pays nothing.
The charge is still computed so a receipt can show the value given, e.g. 0 (1 back). Turning a class off is a settings field, not a deploy.
LLM passthrough is never comped — the 1.5 margin is where the platform's margin on tokens comes from.
Search provider
A search provider is priced per query, with a recovery floor set in credits. Charging a credit price and charging a multiple of what it costs us are the same setting — set one and the other follows.
Cost per query
$0.01
$150 buys 15,000 searches.
Recovery floor (credits)
5
Credits to break even on the query.
Charge (credits)
0
What the user pays.
or × the floor
0
The floor multiplied by the charge.
Absorbed — recorded as cost with no recovery.
What these settings do
Recomputed live from the values above, before anything is saved.
recovery = provider cost ÷ $0.002 · charged = recovery × 1.5, rounded up to a whole credit — roughly 750 credits per dollar of provider cost.
| Scenario | Provider cost | Recovery | Charged | Est. ceiling |
|---|---|---|---|---|
One SerpAPI search $150 buys 15,000 searches | $0.0100 | 5 | 8 | 8 |
$2.00 of LLM usage provider cost | $2.00 | 1000 | 1500 | 1500 |
gpt-4.1-mini · 1k in / 500 out gpt-4.1-mini rates | $0.001200 | 0.6 | 1 | 1 |
claude-opus-4 · 10k in / 2k out claude-opus-4 rates | $0.0400 | 20 | 30 | 30 |
Embedding · 1k tokens text-embedding-3-small · comped in full | $0.000020 | 0.01 | 0 (1 back) | 1 |
Search is shown at the 1.5 floor. Whether search is charged or absorbed is a provider setting — an absorbed call is recorded as cost with no recovery.
Fixed Costs
Base control plane call
Includes:
- -Auth, materialization, policy, cadence and receipts
- -Never waived — every call that crosses the gateway pays it
- -Deterministic tools simply add nothing on top of it
Discovery engine
Includes:
- -2-credit premium on top of the 1-credit base
- -Semantic search across tool surface
- -Prolog planning and path optimization
- -Type inference and adapter generation
- -CTC safety verification
- -Pareto-optimal route selection
Skill minting
Includes:
- -Extracts parameterized workflow
- -Generates reusable MCP tool
- -Caches plan and adapters
- -Creates organizational asset
Multi-agent coordination
Includes:
- -Agent bidding protocol
- -Plan comparison
- -Optimal agent selection
Variable Costs (Cost Recovery + Margin)
LLM calls
- -Formula: recovery = provider cost ÷ $0.002, then charged = ceil(recovery × 1.5) — always a whole credit
- -Rule of thumb: charged ≈ provider cost × 750, rounded up
- -Rates come from the same model dataset the LLM Cost Calculator uses
- -LLM passthrough is never comped — the 1.5 margin is where the platform's margin on tokens comes from
Embeddings
- -Comped in full: the charge is computed, then given back — net 0
- -The receipt still shows the value given, e.g. 0 (1 back)
- -The gateway base still applies to the call
External APIs (SERP, etc.)
- -Charged either as a credit price or as a multiple of what it costs us — the two are the same setting
- -Absorbed calls are recorded as cost with no recovery
BYOK (Bring Your Own Key)
- -LLM costs: 0 credits — there is no passthrough to bill
- -The gateway base and any discovery premium still apply
- -Example: the discovery workflow below drops from 6 credits to 3
Caching & Optimization
- Cached embeddings: 0 credits — embeddings are comped in full
- Cached LLM responses: 1 credit — the gateway base for the call that crosses it
- Cached discovery plans: 1 credit — the premium is skipped, the base is never waived
- Compiled skill execution: 1 credit plus the model cost of the run
Approx. costs provided in estimate and actual costs shown in receipt