For developers building AI agents and autonomous pipelines

Let agents run.
Not your bill.

S3-compatible storage at $4.99/TB flat. No per-PUT fees, no per-GET fees, no egress. Agent loops run at full speed without a per-call counter.

Start for free Talk to an expert

No credit card required · No per-request fees · Connects in minutes

Agents are high-frequency by design. Per-call pricing taxes that frequency.

Every agent step writes state. Every context-aware turn reads memory. Every tool call produces output. Storage that charges per operation makes agents expensive to run — and the cost scales with capability, not with data volume.

The write counter

Every state write is a metered event on S3.

AWS S3 charges $0.005 per 1,000 PUTs. An agent writing state on every step, across hundreds of concurrent runs, generates millions of PUTs per day. At scale, the request bill exceeds the storage bill.

The retrieval tax

Reading context and memory charges egress.

Agents that retrieve context, load memory, or read prior outputs pay $0.09/GB in egress on every read from AWS. The more context-aware the agent, the more it reads — and the more the bill grows.

The framework lock

Purpose-built AI storage charges per query.

Managed vector databases and agent memory platforms charge per API call. Teams building high-frequency agents find that per-query pricing makes the storage layer the dominant cost — not the LLM.

Same boto3. Zero per-call counter.

Any S3-compatible tool your agent already uses connects with an endpoint change. Writes, reads, and lists are all included in flat storage.

agent_storage.py

import boto3, os

# Flat-rate storage — no per-call counter
s3 = boto3.client(
    "s3",
    endpoint_url="https://eu-west-1.s3.fil.one",
    aws_access_key_id=os.environ["FIL_ACCESS_KEY"],
    aws_secret_access_key=os.environ["FIL_SECRET_KEY"],
    region_name="eu-west-1",
)

# Write state on every agent step — $0 per PUT
s3.put_object(
    Bucket="agent-state",
    Key=f"agents/{agent_id}/step-{step}.json",
    Body=json.dumps(state),
)

# Read context on every turn — $0 egress
ctx = s3.get_object(
    Bucket="agent-state",
    Key=f"memory/{agent_id}/context.json",
)["Body"].read()

# High-frequency tool output — still $0 per call
s3.put_object(
    Bucket="agent-state",
    Key=f"outputs/{run_id}/{tool}/{ts}.json",
    Body=json.dumps(tool_output),
)
# 10 million calls today: $0 in request fees

AWS S3 Standard$0.005 / 1K PUTs$5,000

Google Cloud Storage$0.05 / 10K ops$5,000

Azure Blob Storage$0.055 / 10K writes$5,500

Wasabi$0 per request$0

Backblaze B2$0 per request$0

Fil One$0 per request$0

Public US rate cards, Q2 2026. Request fees only — storage and egress are separate on metered tiers and zero on Fil One. AWS: 1,000,000,000 / 1,000 × $0.005 = $5,000. Google: $0.05/10K × 1B = $5,000. Azure: $0.055/10K × 1B = $5,500.

Storage cost that scales with data, not with calls.

The only change is the endpoint. Call frequency stops being a billing variable.

No per-request fees

PUT, GET, LIST, HEAD — all included in flat storage. Agents that write state every turn and read context on every call pay $0 in request fees.

No egress on context reads

Retrieving memory, loading corpus chunks, and reading prior outputs cost $0 in egress. Agent loops that read frequently pay only for what they store.

S3-compatible — existing tools work

boto3, @aws-sdk/client-s3, LangChain S3 loaders, LlamaIndex — any tool that reads or writes S3 connects with an endpoint change. No new SDK.

Cost that scales with data, not calls

$4.99/TB flat. An agent that makes 10 million calls a day but stores 1 TB pays $4.99/month. Call frequency is not a billing input.

One rate. $4.99/TB/month.

Storage. That is the whole bill. Run agents at any frequency — the invoice is determined by bytes stored, not calls made.

Start for free Talk to an expert

No credit card required · No per-request fees · Connects in minutes