BYO compute · access gates · Litheum microtx
Fine-tune open models on your IP.Sell access. Keep the weights.
Specialist hosts meter gated API access on their own GPUs. Every call settles on Litheum.
Agent volume · live ledger
Open trail- $0.006764aegis/pop-rpc-scorerseed-…-burn
- $0.009572northstar/microlend-underwriterseed-…er-15
- $0.001580deepseek/deepseek-r1seed-…r1-14
- $0.000184qwen/qwen3-32bseed-…2b-13
Call path
What happens on every completion
0x0056…live5% rail · 95% hostFor buyers / enterprises
Control the spend. Keep the trail.
For hosts / OSS
Keep the weights. Get paid per token.
Three rails
Bring a model, compute, or both.
Stock, fine-tune, or specialist. Passport shows the class. You keep the weights.
Bring GPUs, join the pool, or let the buyer use theirs. Listing compute is free.
Open API, or approve / deny with competitor blocks. Access is not a fee.
How a call works
Four steps from private fine-tune to settled microtransaction. Built so a Litheum or enterprise buyer can understand the loop in under a minute.
One path. Unequal steps. No generic SaaS grid.
- 01
Host lists a model
Point Cautus Intelligence at your OpenAI-compatible endpoint. Publish USP, base weights, fine-tune story, and pricing. Weights never leave your GPUs.
- 02
Gate who can call it
Open API, approval queue, or allowlist. Block competitor email domains so rivals cannot meter your specialist fine-tune.
- 03
Buyer gets one mg_ key
Browse live listings, request access if gated, then call any allowed slug with a standard OpenAI client.
- 04
Litheum settles every call
Usage + a three-way Litheum settlement (platform / model / compute) append to the ledger. One rail take, 2.5–6%. Microtransactions stay economical at agent volume.
Live marketplace
Frontier routes and gated specialists with online status.
PoP RPC Scorer
PoP-aware RPC scoring trained on Litheum runner traces, not generic uptime pings.
Kimi K2.5
FrontierMoonshot AI
Moonshot AI long-context model. Popular base for specialist fine-tunes on Cautus Intelligence.
Hosted compute · global · base kimi-k2.5
DeepSeek R1
FrontierDeepSeek
Open-weight reasoning specialist with competitive math and code performance.
Hosted compute · global · base deepseek-r1
Llama 4 Maverick
FrontierMeta
Meta open-weight workhorse for private deployments and high-throughput apps.
Hosted compute · global · base llama-4-maverick
For model hosts
List once. Serve from your stack.
- ✓Weights stay on your GPUs. Cautus Intelligence only routes and meters.
- ✓Heartbeats publish online / offline / degraded to buyers.
- ✓Hate who uses your models: approval queues + competitor domains.
For API buyers
One key. Niche specialists included.
- →Browse live models with pricing, origin, and access rules.
- →Request access when a host gates the endpoint.
- →Drop-in OpenAI path: same SDK, marketplace slug as model id.
Why Litheum for microtransactions?
A single completion can cost fractions of a cent. Traditional monthly invoices lose the audit trail. Litheum targets high-throughput micro-finance so Cautus Intelligence can log usage, fee split, and host payouts without gas becoming the product.
Opaque rollup. Hard to audit per agent or per buyer.
Every call is a microtx hosts and buyers can verify.
Hate who uses your models
Competitor domain blocks are a real control, not a paragraph. Add a domain and the gate rejects that buyer email at request and chat time.
Demo control. Live listings enforce these domains on chat and access requests.
BYO compute
Cautus Intelligence routes and meters. You keep the hardware, the VPC, and the uptime story buyers see on every listing card.
2.5–6% rail
One take per call, by contribution mode. Not a landlord cut. Visible on Litheum.
Where Cautus Intelligence sits
Not a weight hub. Not a generic router. Host-owned specialist access with policy and microtx settlement.
| Hugging Face | OpenRouter-style | Cautus | |
|---|---|---|---|
| What is sold | Weights / repos | Aggregated APIs | Host-owned specialist access |
| Weights leave host? | Often yes | N/A | No |
| Access policy | Repo licenses | Key to router | Open / approval / allowlist + competitor domains |
| Settlement | Off-platform | Provider billing | Per-token Litheum, 2.5–6% rail |
Open weights + private IP is the next model economy
Teams take open-source bases, continue train on proprietary corpora, and resell capability as API access without shipping weights. That needs gating, metering, competitor exclusion, and auditable payouts.
Cautus Intelligence is that marketplace. Litheum is the microtransaction rail. Non-chain features still matter: BYO compute, approval queues, domain blocks, and OpenAI-compatible keys.
Read the market thesisDrop-in OpenAI client
Same SDK. Marketplace slug as model id.
curl https://api.cautusintelligence.com/v1/chat/completions \
-H "Authorization: Bearer mg_..." \
-H "Content-Type: application/json" \
-d '{
"model": "aegis/pop-rpc-scorer",
"messages": [{"role":"user","content":"Score this runner's last 50 critical RPCs for PoP credit"}]
}'Show the loop in five minutes
One-click demo session, issue a key, call a gated specialist, open the Litheum receipt. Built for the Litheum team walkthrough.