Storage architecture
Omni- PostgreSQL only¹
- Onyx
- PostgreSQL + OpenSearch + Redis + MinIO/S3¹
The Postgres-native alternative to Onyx
Omni uses PostgreSQL as its only storage system for application data, BM25 and vector search. Permission syncing from connected apps is included in the open-source core.
One PostgreSQL data plane
Application data, documents, BM25 and vector index in one system.
Permission inheritance included in OSS
Source-system access controls are part of the Apache 2.0 open-source core.
4 vCPU / 8 GB recommended footprint
Omni can operate comparable workloads with lower infrastructure requirements.
At a glance
Why choose Omni over Onyx?
| Feature | Omni | Onyx |
|---|---|---|
| Storage architecture | PostgreSQL only¹ | PostgreSQL + OpenSearch + Redis + MinIO/S3¹ |
| Permission-aware search | Included in open-source core² | Commercial edition² |
| Hybrid retrieval | ParadeDB BM25 + pgvector | OpenSearch + knowledge graphs |
| Connectors | 20+ with standalone SDKs³ | 40+ built in³ |
| Scheduled agents | Native cron and interval agents | Not available. Workflows coming soon⁴ |
| Deep Research | Coming soon | Available today |
| Open-source licensing | Apache 2.0 open-source core | MIT Community Edition + commercial EE⁵ |
Permission-aware search
- Omni
- Included in open-source core²
- Onyx
- Commercial edition²
Hybrid retrieval
- Omni
- ParadeDB BM25 + pgvector
- Onyx
- OpenSearch + knowledge graphs
Connectors
- Omni
- 20+ with standalone SDKs³
- Onyx
- 40+ built in³
Scheduled agents
- Omni
- Native cron and interval agents
- Onyx
- Not available. Workflows coming soon⁴
Deep Research
- Omni
- Coming soon
- Onyx
- Available today
Open-source licensing
- Omni
- Apache 2.0 open-source core
- Onyx
- MIT Community Edition + commercial EE⁵
- 1 Omni consolidates application data, documents, BM25 search, vectors in PostgreSQL. Current Onyx releases use OpenSearch; older releases used Vespa, with v3 operating both during migration. Onyx repository
- 2 Both products support permission-aware retrieval. Omni includes source-system ACL inheritance in its open-source core, while Onyx requires a paid edition for document permission inheritance. Onyx connector and permissions docs
- 3 Onyx offers more prebuilt connectors. Omni differentiates through standalone Python, TypeScript, and Rust SDKs, with each connector deployed as an isolated service. Onyx connectors are primarily implemented as in-repository Python classes, with an ingestion API for custom sources. Onyx connector docs
- 4 Omni agents can run automatically on cron or interval schedules. Onyx offers interactive custom agents, while its time-based Workflows feature is currently documented as coming soon. Onyx Workflows documentation
- 5 Onyx Community Edition is MIT licensed, but features under its
eedirectories – including permission functionality – require a commercial production license. Omni’s enterprise features are maintained separately. Onyx license
Omni vs Onyx: Which Open-Source AI Agent Should You Self-Host?
This comparison focuses on the practical trade-offs between the two open-source editions: permission inheritance, the systems underneath retrieval, connectors, deployment, and the cost of running them.
01 / Open-source editions
Open-source editions compared
Omni’s open-source philosophy is to include all core features in the Apache 2.0 licensed repository. This includes important features like permission inheritance, which is not open-source in Onyx and requires purchasing an enterprise license.
Permission inheritance is the mechanism by which systems like Omni and Onyx ensure that users can only access the data that they already have access to in the source systems. E.g., a user should be able to access a certain Google Doc in Omni only if they have read access to it on Google Docs itself.
Onyx gates this feature behind a commercial license, which means that teams that want to deploy the open-source version to test drive the permission handling won’t be able to do so without purchasing a license.
02 / Context and connectors
Connectors are part of the operating model
Omni’s connector model combines a core catalog with a flexible SDK path. See Omni’s connector catalog or use its standalone Python, TypeScript, and Rust SDKs to build custom integrations. Omni provides connectors to most core applications commonly used, and we build additional connectors that might be valuable to customers once a commercial use agreement is made. The guiding philosophy with Omni is that we don’t let the connectors come in the way of customer success in our implementations. Onyx has more prebuilt connectors; see Onyx’s connector documentation for its supported sources and permission model.
For more detail on Omni’s retrieval approach and results, read our enterprise RAG benchmark.
03 / Architecture
Architecture: Postgres-first vs multi-service
Omni uses PostgreSQL as its primary data plane for application data, documents, BM25 and vector search, and task queues. By contrast, a standard Onyx deployment uses PostgreSQL alongside OpenSearch, Redis, and MinIO/S3. The recommended node size for Onyx is 8 vCPUs and 16GB memory; Omni can operate comparable workloads with 4 vCPUs and 8GB memory.
Omni’s Postgres-first design minimizes long-term operational burden. All core Omni services are written in Rust, which means the typical Omni deployment is very efficient on resources. Unlike Onyx’s OpenSearch-based architecture, there is no separate JVM heap to size and tune, search cluster to maintain, or additional data store to back up and recover.
The best way to evaluate the production footprint is to deploy Omni with Docker Compose; it only takes about 10 minutes. Omni’s AWS Terraform deployment is available for teams that need a cloud deployment path.
04 / Deployment and operations
Running each platform in production
Both platforms can be run entirely in infrastructure you own and operate (public/private cloud or on-premises). Omni supports deployment with Docker Compose and AWS/GCP Terraform, while Onyx also supports Kubernetes and Helm, in addition to these. See the Onyx deployment guide for its current installation paths. Omni has a lighter operational footprint, with no OpenSearch cluster or JVM heap to monitor and tune. Also, Onyx has a managed cloud available today, whereas Omni’s cloud offering will be available in the coming weeks.
05 / Pricing
Omni vs Onyx: Pricing
Both platforms offer $0 self-hosted community editions. As described earlier, Omni includes permission inheritance in its community edition, while Onyx includes it only in its paid edition. Neither community edition includes enterprise features like SSO, SCIM, audit logs, or an enterprise support SLA. The cards below show each platform’s paid tier; infrastructure and operations are separate. See Onyx’s current pricing and feature tiers for the hosted and enterprise details.
06 / Choosing between them
Which is the better fit for your team?
Choose Omni if you value:
- Permission inheritance in the Apache 2.0 open-source core
- One PostgreSQL data plane, fewer stateful dependencies, and lower recommended infrastructure requirements
- Postgres-first hybrid retrieval and native scheduled autonomous agents
- 20+ native connectors available today
- Each connector runs in its own Docker container, with SDKs in Python, TypeScript, and Rust
- Self-hosted deployment with Docker Compose and Terraform
- Open-source control over data, models, and operations
Choose Onyx if you need:
- Permission inheritance in the paid enterprise edition; MIT Community Edition for the open-source core
- PostgreSQL alongside OpenSearch, Redis, and MinIO/S3 for a broader multi-service stack
- Deep Research and knowledge-graph-enhanced retrieval
- 40+ prebuilt connectors
- Custom sources through an ingestion API and repository-based connectors
- Kubernetes/Helm deployment
- Desktop, browser-extension, or Slack-bot access, plus voice/image generation and Craft app creation
1 Examples currently include Asana, Airtable, Box, Dropbox, GitLab, Zendesk, and Zoom. Connector availability changes frequently; check both catalogues for the latest coverage.
2 Deep Research and interactive app and dashboard generation are on Omni’s roadmap.
07 / FAQs
Frequently asked questions
What is included in Omni’s open-source edition?
Omni includes all core features in the Apache 2.0 licensed repository, including permission inheritance. The supplied comparison describes permission inheritance as a commercial feature in Onyx.
What does permission inheritance mean?
Permission inheritance ensures that users can only access data they already have access to in the source systems. For example, a user should access a Google Doc in Omni only if they have read access to it in Google Docs itself.
How do Omni and Onyx differ architecturally?
Omni uses PostgreSQL for the search index, including BM25 and vector search, and for all app data. A standard Onyx deployment requires PostgreSQL, OpenSearch, and MinIO/S3; the comparison table also identifies Redis in its storage architecture.
Can both products be self-hosted?
Yes. Both platforms can run entirely in infrastructure you own and operate, including public or private cloud and on-premises environments. Omni supports Docker Compose and AWS/GCP Terraform; Onyx also supports Kubernetes and Helm.
How does self-hosted pricing compare?
Both platforms offer free self-hosted community editions. Onyx Cloud Business costs $20 per user/month billed annually ($25 billed monthly), while the upcoming Omni Cloud is priced at $10 per user/month billed annually ($15 billed monthly).
Ready to evaluate?
Deploy Omni in your environment
View Omni on GitHub or contact us to book a demo.