Ditto is a persistent-memory AI assistant: it builds a personal knowledge graph out of every conversation, so an agent can pick up context it would otherwise forget between sessions. That memory is built from real files, images, documents, and soon video, plus a growing stream of data generated by Ditto's own AI agents, which makes storage a genuinely load-bearing part of the product rather than a background utility.
To run that storage layer, Ditto uses Hippius, a distributed cloud platform built as a verifiable alternative to providers like AWS and Backblaze. Ditto moved off Backblaze earlier this year, and this case study covers why, how Hippius fits into Ditto's stack today, and why the team has started routing its AI agent traces there too.
Why Ditto Moved Off Backblaze
Ditto's product depends on remembering things reliably. Every image, document, and eventually video a user shares becomes part of that persistent memory, which means the storage layer underneath it has to be dependable at a file level, not just fast at a database level. Before Hippius, Ditto ran this on Backblaze.
Backblaze itself wasn't the problem, although its B2 storage isn't a fully faithful implementation of the S3 API, which was a compatibility gap for a team used to working against the standard S3 spec.
However, this issue alone wasn't big enough to force the change. What actually moved Ditto off Backblaze was a combination of straightforward cost savings and a strategic reason: Ditto also mines on Bittensor (its DittoBench benchmark runs as Subnet 118), and switching to a Bittensor-native storage provider kept that spend inside the network its own infrastructure already depends on, rather than sending it to a centralized provider with no connection to the ecosystem.
That's a fairly common path among teams evaluating backblaze alternatives: the incumbent works well enough, but cost and a better-aligned pricing model make switching worthwhile.
"We switched from Backblaze to Hippius for two reasons: the cost savings, and it lets Ditto keep our storage spend inside the Bittensor network our own infrastructure already runs on, instead of sending it to an unrelated centralized provider."
— Peyton Spencer, CEO at Ditto
User Files and Agent Traces, Both Running on Hippius
Hippius now handles two distinct workloads for Ditto.
The first is user files: images and documents today, with video support planned next. This is the storage that sits directly under Ditto's knowledge graph, since a memory that references a photo or a PDF needs that file to actually be there and retrievable whenever the agent pulls it back up.
The second is newer: agent traces produced by the Ditto subnet. As Ditto's agents run tasks, they generate their own operational data, and the team has started storing all of it on Hippius rather than splitting it across providers. That data already runs into the tens of gigabytes, and it's the kind of workload built to keep growing, since every agent run adds more of it.
Both workloads run through Hippius's S3-compatible object storage, which behaves like AWS S3 from a developer's point of view: same API surface, same tooling. That's a meaningfully closer match to the standard S3 spec than Ditto had with Backblaze, so code written against the S3 API works against Hippius without the compatibility gaps the team had previously run into. Every storage operation is also recorded on Hippius's own blockchain, checkable on the public Hipstats explorer, which extends the same transparency Ditto already applies to its own agents' memory retrieval into the storage layer underneath them.
Why Hippius Was the Right Storage Solution
A few things made Hippius the right move for Ditto specifically, beyond the general appeal of distributed storage:
- Lower cost at current scale. Ditto's user storage is still well under a terabyte, and running that volume on Hippius costs next to nothing, which matters for a team scaling a product rather than optimizing an already-large storage bill.
- Full S3 compatibility. Where Backblaze's B2 diverged from the standard S3 spec in places, Hippius's object storage matches it closely enough that Ditto hasn't run into equivalent friction.
- Bittensor-aligned revenue. Ditto already runs infrastructure on Bittensor through DittoBench. Storing on a Bittensor-native provider like Hippius keeps that spend inside the same network rather than routing it to an unrelated centralized vendor. For a team already building on Bittensor, this kind of object storage for AI work is a natural extension rather than an added dependency.
- Clean headroom for agent data. Agent traces are a newer, faster-growing workload than user files, and Ditto has been able to scale that use case on Hippius without hitting friction, which matters given how quickly agentic AI systems can generate operational data.
Notably, Ditto didn't switch because Backblaze failed at anything. It switched because Hippius did the same job for less, with a cleaner API match and a storage bill that supports infrastructure Ditto already depends on.
Storage Costs Near Zero, Even as Agent Traces Scale
Ditto now runs all of its user file storage and its agent trace data on Hippius, with no compatibility issues reported since the switch.
"Our storage costs on Hippius are nearly zero. We've moved every trace our AI agents generate onto Hippius, and it keeps scaling cleanly without any issues and minimal cost."
— Peyton Spencer, CEO at Ditto
For a persistent-memory AI assistant, that combination, low cost, full S3 compatibility, and Bittensor-aligned infrastructure, is exactly what a storage layer supporting both user memory and agent activity needs to hold up as usage grows.
If your own team is weighing Backblaze, AWS, or another centralized provider against something built for AI-scale storage, see current Hippius pricing and check what an S3-compatible switch would look like for your workload.
About Ditto
Ditto is a persistent-memory AI assistant that builds a personal knowledge graph out of every conversation, so agents can act on context that would otherwise be lost between sessions. It connects across chat platforms, supports agentic workflows through sub-agents and workflows, and runs its own benchmark, DittoBench, as Bittensor Subnet 118. Ditto is built by Omni Aura.
Website | Blog | Docs | X | GitHub
About Hippius
Hippius is a distributed cloud platform built to give users a verifiable alternative to providers like AWS and Google Cloud. Files are encrypted and distributed across an independent network rather than held in a single provider's data center. Every storage claim, payment, and miner action is recorded on Hippius's own blockchain, so instead of asking users to trust the platform, Hippius lets them check it for themselves. On top of that foundation sits a full product suite: personal cloud storage, S3-compatible object storage with no egress fees, confidential computing on secure hardware, and a container and model registry for AI. Hippius was built by The Nerve Lab.
Website | Discord | Twitter | Docs | Blog | Medium | Forum | Github