Skip to content
← Marketplace

First-class passport · hashes on the ledger, artifacts stay off-chain.

A

Agent Settlement

OnlineFine-tuned

Northstar Policy AI · Chris Park

northstar/agent-settlement

Why this model

Routes DeFAI inference buys so Litheum per-token receipts stay economical.

DeFAI settlement helper: routes agent inference buys so Litheum per-token receipts stay economical at agent volume.

Northstar continued training Llama 4 Maverick on Cautus Intelligence usage receipts and agent loop traces. Suggests batching, model fallbacks, and settlement checks so AI agents can buy inference continuously without fee overhead dominating unit economics.

Best for
batching advicemodel fallbackssettlement checksagent loops
Input $/1M
$0.9
Output $/1M
$2.8
Context
128,000
Origin
Domain fine-tune
Model class
Fine-tune
Contribution
Model only (matched)
Compute source
Compute pool / third party
License
Commercial
Base model
meta/llama-4-maverick
Compute region
us-east-1
Requests
1
Host wallet
0xNORTHS…0000000E
Hosted compute

Weights and inference stay on the host. Cautus Intelligence meters the gateway; buyers never receive the raw endpoint credentials.

Autoscale A10G fleet · BYO agent traces

Endpoint registered (proxied): https://settle.northstar.internal/v1

Last heartbeat 10/11/2026, 4:19:10 PM

Data and training story

Cautus Intelligence usage receipts and DeFAI agent loop traces.

Model provenance passport
Hash on-chain · artifacts off-chain

Lineage a buyer can click: base weights, license, fine-tune story, region, and custody. We do not invent eval scores. If no third-party card exists, the listing says so and still pins a card hash.

Base weights
meta/llama-4-maverick
License
Commercial
Origin
Domain fine-tune
Model class
Fine-tune
Contribution
Model only (matched)
Compute source
Compute pool / third party
Region
us-east-1
Weights custody
Weights never left this host GPU / VPC.
Eval / card hash
0xdccd35f0f378b30d37164f2898a9164fc6de4bdcb1d15899
Eval status
No third-party eval published for this listing.
Litheum listing tx
0xa36980bc…2429a9126c
Fine-tune story

Northstar continued training Llama 4 Maverick on Cautus Intelligence usage receipts and agent loop traces. Suggests batching, model fallbacks, and settlement checks so AI agents can buy inference continuously without fee overhead dominating unit economics.

View listing on Litheum
Access control

Mode: Open API

Open API metering. Weights stay on Northstar GPUs.

Open API access

No approval queue. Sign in for credits, create an mg_ key, and call this slug. Competitor domain blocks still apply if the host configured them.

Settlement path

This call → usage tx → settlement tx on Litheum. Platform take on this listing is 6.0% (Model only (matched)). Remainder goes to the model owner and, on matched jobs, the compute provider. Example at $0.9 / $2.8 per 1M: a 1M in + 1M out call costs $3.7.

Open Litheum explorer