Predict
$0.03 / 1K
Stages likely memory before the user asks. Best for voice agents, live assistants, and low-latency agent loops.
Adaptive PQHR@585.9%
Warm staged P500.85ms
Outputhandles
Predict-first memory pipeline
MemFront console
MemFront returns handles first, keeps the likely context hot, and hydrates full memory only when the customer needs it.
Products
Four separately metered layers. Customers can use one layer or the full pipeline.
Memory
$0.10 / 1K
Explicit memory search over compact metadata. Returns ranked handles, previews, scores, and storage keys.
LME retrieval@590.6%
LoCoMo retrieval@585.1%
Storagemetadata
Hydration
$0.05 / 1K + bytes
Fetches raw transcript or file content from Cloudflare R2 after Predict or Memory selects the handle.
Local API P500.49ms
Cold storeR2
Inputstorage_key
Reasoning
$5 BYOK / $40 hosted
Optional answer layer. It uses Memory plus Hydration, then runs an LLM only when the customer asks for final QA.
E2E LLM judge85.2%
ModeBYOK/hosted
Pathpremium
Hosted architecture
1
Customer app logs turns
SDK posts events with an API key.
2
Hot metadata stays in Supabase
Terms, previews, timestamps, handles, usage.
3
Raw memory goes to R2
Full text is fetched only by `/hydrate`.
4
Reasoning is optional
LLM calls happen only on the premium path.
API health
unknown
Connect to your MemFront API endpoint.
Account
none
Paste a bearer key to authenticate.
Billing
unknown
Checkout activates when the API has a Stripe link.
Connection
For hosted Vercel deploys this defaults to same-origin /api. Local development uses http://localhost:8000.
Config
No config loaded.
API playground
Exercise the hosted API exactly like a customer app: log memory, stage predictions, hydrate context, or run Reasoning.
Response
No request yet.
Usage
Per-product telemetry for billing and customer usage reporting.
Predict units
0
Memory units
0
Hydration bytes
0
Reasoning units
0
No usage loaded.
API keys
Admin-only minting uses the configured admin token and returns the raw key once.
Created key
No key created.
Billing
Simple pay-as-you-go packaging. Predict, Memory, and Hydration stay cheap; Reasoning is the premium LLM path.
Predict
$0.03 / 1K staged prediction requests
Memory
$0.10 / 1K retrieval requests
Hydration
$0.05 / 1K R2 fetches plus returned bytes
Reasoning BYOK
$5 / 1K orchestration calls
Reasoning Hosted
$40 / 1K answer-generation calls
No checkout request yet.