# Fil One: Full Reference for AI Agents and Developers

> Fil One is the storage layer for AI agents. S3-compatible object storage built on
> Filecoin with no egress penalties, no per-request charges, and flat pricing at
> $5.99/TB/month. Free 30-day trial, no credit card required.

---

## What is Fil One?

Fil One provides S3-compatible object storage built on Filecoin. Each region is run
by an independent storage operator: Europe (France) as `eu-west-1`, the default, and
US East (Michigan) as `us-east-1`. More regions are on the way.

Storage is designed for 11 nines of durability. Object lock (governance or
compliance retention, set at bucket creation) and versioning protect data from
being overwritten or deleted.

Reads are free: there are no egress penalties and no per-request charges, so agent
workloads that read data often pay only for what they store.

---

## AI agent use cases

### 1. RAG document corpus and vector store backing

Store the raw or pre-processed documents that back a retrieval-augmented generation
pipeline. Fil One handles the high-write ingestion phase (chunking, embedding
creation) and the high-read retrieval phase (fetching source documents for context
injection) with no per-request charges.

```python
import os, boto3

s3 = boto3.client(
    "s3",
    endpoint_url="https://s3.eu-west-1.filonecontent.com",
    aws_access_key_id=os.environ["FIL_ONE_ACCESS_KEY"],
    aws_secret_access_key=os.environ["FIL_ONE_SECRET_KEY"],
    region_name="eu-west-1",
)

# Ingest a document chunk
s3.put_object(
    Bucket="rag-corpus",
    Key=f"chunks/{doc_id}/{chunk_index}.txt",
    Body=chunk_text.encode(),
    Metadata={"embedding-model": "text-embedding-3-small", "doc-id": doc_id},
)

# Retrieve source for citation
obj = s3.get_object(Bucket="rag-corpus", Key=f"chunks/{doc_id}/{chunk_index}.txt")
source_text = obj["Body"].read().decode()
```

### 2. Agent memory and persistent state

Persist episodic memory, conversation history, task queues, scratchpads, and
structured world-state between agent runs. Fil One is a durable key-value store
accessible from any process with S3 credentials, with no database required.

```python
import os, json, boto3

class AgentMemory:
    def __init__(self, bucket: str, agent_id: str):
        self.s3 = boto3.client(
            "s3",
            endpoint_url="https://s3.eu-west-1.filonecontent.com",
            aws_access_key_id=os.environ["FIL_ONE_ACCESS_KEY"],
            aws_secret_access_key=os.environ["FIL_ONE_SECRET_KEY"],
            region_name="eu-west-1",
        )
        self.bucket = bucket
        self.prefix = f"agents/{agent_id}/memory"

    def save(self, key: str, data: dict):
        self.s3.put_object(
            Bucket=self.bucket,
            Key=f"{self.prefix}/{key}.json",
            Body=json.dumps(data).encode(),
        )

    def load(self, key: str) -> dict | None:
        try:
            obj = self.s3.get_object(
                Bucket=self.bucket, Key=f"{self.prefix}/{key}.json"
            )
            return json.loads(obj["Body"].read())
        except self.s3.exceptions.NoSuchKey:
            return None
```

### 3. Model artifact hosting

Store and serve fine-tuned model weights, GGUF/ONNX files, LoRA adapters, tokenizer
configs, and quantized checkpoints. Generate pre-signed URLs so agent runtimes or
inference servers can pull artifacts directly without exposing credentials.

```python
# Upload a fine-tuned model checkpoint
s3.upload_file(
    "model-finetuned-v2.gguf",
    "model-artifacts",
    "runs/v2/model.gguf",
)

# Generate a pre-signed URL valid for 1 hour (for inference server pull)
url = s3.generate_presigned_url(
    "get_object",
    Params={"Bucket": "model-artifacts", "Key": "runs/v2/model.gguf"},
    ExpiresIn=3600,
)
```

### 4. Inference trace archival and evaluation datasets

Archive prompts, completions, tool call traces, and structured outputs for offline
evaluation, fine-tuning data collection, or compliance logging. Write as JSONL and
read back in bulk with no egress cost.

```python
import json

def archive_trace(trace: dict, run_id: str, step: int):
    key = f"traces/{run_id}/{step:06d}.json"
    s3.put_object(
        Bucket="inference-logs",
        Key=key,
        Body=json.dumps(trace).encode(),
        ContentType="application/json",
    )

def load_traces_for_eval(run_id: str) -> list[dict]:
    paginator = s3.get_paginator("list_objects_v2")
    traces = []
    for page in paginator.paginate(Bucket="inference-logs", Prefix=f"traces/{run_id}/"):
        for obj in page.get("Contents", []):
            body = s3.get_object(Bucket="inference-logs", Key=obj["Key"])["Body"]
            traces.append(json.loads(body.read()))
    return traces
```

### 5. Training and evaluation datasets

Store JSONL, Parquet, CSV, or HDF5 datasets and stream them back to training jobs.
No egress penalty means iterating over datasets repeatedly costs only the storage price.

```python
import os
import pyarrow.parquet as pq
import pyarrow.fs as pafs

# PyArrow native S3 access (pass custom endpoint via S3FileSystem)
fs = pafs.S3FileSystem(
    endpoint_override="s3.eu-west-1.filonecontent.com",
    access_key=os.environ["FIL_ONE_ACCESS_KEY"],
    secret_key=os.environ["FIL_ONE_SECRET_KEY"],
    scheme="https",
)

table = pq.read_table("my-datasets/train/sft-v3.parquet", filesystem=fs)
```

---

## Connection reference

| Parameter | Value |
|-----------|-------|
| Endpoint URL | `https://s3.eu-west-1.filonecontent.com` (default region) |
| Region | `eu-west-1` (Europe, France) |
| Other regions | `us-east-1` (US East, Michigan) at `https://s3.us-east-1.filonecontent.com` |
| Auth | AWS Signature V4 (Access Key ID + Secret Access Key) |
| Protocol | HTTPS only; HTTP is rejected |
| Addressing | Path-style required; virtual-hosted-style is not supported |

Fil One is compatible with any S3-compatible client or SDK:
- **Python:** boto3, s3fs, fsspec, PyArrow S3FileSystem
- **Node.js/TypeScript:** @aws-sdk/client-s3, @aws-sdk/lib-storage (multipart)
- **Go:** aws-sdk-go-v2
- **Rust:** aws-sdk-rust, object_store
- **CLI:** AWS CLI v2, rclone, s5cmd
- **Terraform:** aws provider (s3 resources with custom endpoint)
- **LangChain:** S3FileLoader, S3DirectoryLoader

### Environment variable pattern (recommended)

```bash
export FIL_ONE_ENDPOINT=https://s3.eu-west-1.filonecontent.com
export FIL_ONE_REGION=eu-west-1
export FIL_ONE_ACCESS_KEY=your_access_key_id
export FIL_ONE_SECRET_KEY=your_secret_access_key
```

---

## Data protection

- **Durability:** designed for 11 nines
- **Object lock:** governance or compliance retention, enabled when the bucket is
  created. Useful for keeping inference logs and evaluation datasets unaltered for
  a fixed period
- **Versioning:** keep every version of an object, so an overwritten checkpoint or
  dataset can be restored
- **Encryption:** data is encrypted at rest, with keys held by the regional storage
  operator. To hold your own keys, encrypt client-side before upload

---

## Pricing

| Plan | Price | Notes |
|------|-------|-------|
| Pay as you go | $5.99/TB/month | No egress penalties (subject to reasonable use), no API request or retrieval fees; $5.99 monthly minimum |
| Free trial | $0 | 30 days, 1 TB storage and 2 TB egress included, no credit card required; upgrade to keep downloading after the egress allowance |
| Reserved capacity | Custom | 1, 3 or 5-year terms; capacity assurance; deployment SLAs; guided migration and invoicing |

No egress penalties means agent workloads that read data frequently (RAG pipelines,
evaluation loops, inference servers pulling weights) pay only for storage, not for
bandwidth.

For comparison, published list rates (checked 2026-09-23):

| Provider | Storage | Egress |
|----------|---------|--------|
| Fil One | $5.99/TB/month | $0 |
| Wasabi | $7.99/TB/month | $0 |
| Backblaze B2 | $6.95/TB/month | Free up to 3x stored data, then $10/TB |
| Cloudflare R2 | $15/TB/month | $0 |
| AWS S3 Standard (eu-west-1) | $23/TB/month (first 50 TB, tiered) | $90/TB (first 10 TB, tiered; first 100 GB free) |

---

## S3 API compatibility

Fil One implements the core of the S3 REST API. Supported in all regions:

- Bucket: ListBuckets
- Object: PutObject, GetObject (including range requests and `VersionId`),
  DeleteObject, HeadObject, ListObjectsV2, ListObjectVersions
- Multipart: CreateMultipartUpload, UploadPart, CompleteMultipartUpload,
  AbortMultipartUpload, ListMultipartUploads
- Presigned URLs: presigned GetObject and PutObject

Region-dependent:

- CreateBucket and DeleteBucket work in `us-east-1`. In `eu-west-1`, create and
  delete buckets from the dashboard
- CopyObject, object retention, and legal hold are confirmed in `us-east-1`

The full, current list is in the
[S3 compatibility reference](https://docs.fil.one/reference/s3-compatibility).

---

## Pages and documentation

- [Homepage](https://www.fil.one): product overview, feature comparison, pricing, FAQ
- [Pricing](https://www.fil.one/pricing): flat per-TB rate and a cost calculator
  comparing Fil One with AWS S3, Wasabi, Backblaze B2, and Cloudflare R2
- [Bring a workload](https://www.fil.one/workloads): AI training and models, data lakes
  and ETL, backup and archive, media, and security logs
- [Host a zone](https://www.fil.one/host): S3 object storage installed in a GPU
  cloud's own facility, sold under the partner's brand
- [Partners](https://www.fil.one/partners): channel, technology, and MSP partner programs
- [About](https://www.fil.one/about): the team and company behind Fil One
- [Documentation](https://docs.fil.one): quickstart, SDK examples, API reference,
  security, billing, limits
- [Talk to sales](https://www.fil.one/contact-sales): enterprise and high-volume
  inquiries; also at sales@fil.one
- [Privacy Policy](https://www.fil.one/privacy)
- [Terms of Service](https://www.fil.one/terms)