What is OnticFS?
OnticFS is a filesystem where the mount itself is the agent orchestration surface. It allows you to mount a knowledge graph as a POSIX-compliant filesystem. Instead of interacting with a chat interface or a REST API, you interact with typed memory using standard UNIX tools.
- Directories are graphs: Navigate entities via
lsandcd. - Edges are wiki-links: Write
[[depends-on:uuid]]in a text file to wire a graph edge. - Recall is
cat: Read fromby-query/to trigger vector search.
It forces probabilistic AI into auditable, deterministic outputs at the OS boundary.
The Community Edition
Honest disclosure: The Community Edition is a free tier hosted on our own bare metal servers. It is capacity-limited and comes with NO SLA.
We built the Community Edition for hackers, homelabbers, and open-source contributors. It provides access to the core OnticFS capabilities without the overhead of enterprise infrastructure.
If you require SOC2 compliance, guaranteed uptime, or managed deployments in Google Cloud tenants, please refer to the enterprise offering.
Cartridge Ecosystem
OnticFS uses BYOP (Bring Your Own Physics) YAML cartridges to define the ontology of your filesystem. A cartridge specifies primitive subtypes, edge types, wiki-link markers, and automatic MIME routing.
Current community cartridges include:
sys:famix-v1— For software architecture and code analysis.sys:commerce-v1— For product listings, pricing, and merchant data.sys:work-graph-v1— For task dependency and project management.
When you write a recognized marker (e.g., [[price-observed:snapshot-id]]) in a file, the filesystem intercepts the write, validates it against the active cartridge, and wires the corresponding edge in the underlying graph.
Bring Your Own Inference
OnticFS doesn't compete with your model provider — it doesn't care which one you use. We're not in the business of training frontier models or racing anyone's compute budget. We do ontology: reliable, typed grounding for whatever inference you already trust, so the model stays replaceable and the graph underneath it doesn't.
For teams that need privacy guarantees on top of that — prompts and outputs never visible to the inference operator, not even under legal compulsion — the strongest option today is TEE-attested inference secured by decentralized compute networks. Hardware-level attestation (Intel TDX, NVIDIA confidential GPU) spread across an incentivized, permissionless provider set gives you a stronger trust story than any single centralized vendor can: no one operator to subpoena, no one host to compromise.
If crypto-adjacent infrastructure makes you uneasy, we can also just run it for you — a fully hosted inference tier is part of the enterprise offering.
Coming soon: we're finalizing direct relationships with specific TEE-backed inference providers and will name names here once those conversations are done. Ask us in the meantime — hello@onticfabric.io.
Get Involved
The OnticFS kernel is open-source. You can find the source code, report issues, and submit patches on our repository.
git clone https://git.onticfabric.io/onticfs.git # TODO: repo hosting being finalizedTo get early access to the public cartridge registry and request provisioning on the community tier bare metal, join the mailing list below.