Repository navigation
Orama Lake v0.1.0
Released 2026-11-16. The first release of Orama Lake, the sovereign AI stack, published as open source under the MIT licence. This repository holds no product source: a release is one set of component versions that work together, pinned as git submodules. This page lists them, and release.yaml carries the same set as a manifest the installer reads.
Get it
$ git clone --recurse-submodules --branch v0.1.0 https://github.com/oramasearch/lake-platform.gitEvery submodule is fetched over HTTPS, so the clone needs no GitHub account or key. To install the platform, start at docs/install/.
What it pins
| Component | Version | What it is |
|---|---|---|
| lake | v0.9.0 | The workspace: where people, scripts and agents work, through one API served as REST, OpenAPI and MCP |
| zetta | v0.6.0 | The engine: data, SQL, full-text search and machine learning, one per workspace |
| liquid | v0.3.0 | Source control, builds, packages and deploys in one program |
| orama-operator | v0.12.0 | Runs the platform on Kubernetes from resources you declare |
| zetta-tunnel | v0.3.1 | Tunnels, the mutual-TLS mesh and ingress |
| liquify | v0.1.0 | The IDE |
| lake-sqlite | v0.2.0 | lake's SQLite service |
| lake-duckdb | v0.2.0 | lake's DuckDB service |
| lake-polars | v0.1.0 | lake's Polars service |
| lake-builder | v0.1.0 | Builds the images lake deploys |
| app-runtime | v0.2.0 | What the apps people build run on |
| methodology-runtime | v0.1.0 | What machine-learning methodologies run on |
| model-runtime | v0.1.1 | Training and inference for the models |
| agent-runtime | v0.3.1 | What every agent runs on |
| agent-creator | v0.3.1 | Builds the agent you describe |
| agent-application | v0.3.1 | Builds and deploys web and data apps from a description |
| agent-autosuggest | v0.3.1 | Proposes what to build from your situation and your data |
| agent-ask-orama | v0.3.0 | Ask Orama, the assistant on every page of lake |
| agent-project | v0.3.0 | Builds and operates solutions and pipelines |
| agent-experiments | v0.3.0 | AutoML and experiment design |
| agent-eval-generator | v0.3.0 | Writes the test cases agents are measured against |
| agent-project-corpus | v0.3.0 | Training data for agent-project |
| agent-autosuggest-corpus | v0.3.0 | Training data for agent-autosuggest |
| base-agent-0.6b | v0.1.0 | The base model, as an image; its weights are on Hugging Face as OramaSearch/base-agent-0.6b |
| k0s | a pinned commit (db9032ae), with liquid v0.3.0 embedded |
The small Kubernetes distribution, a fork of k0s |
The components group under two bridges, liquid-bridge and zetta-bridge, which pin them; this repository pins the bridges, orama-operator, agent-experiments and model-runtime. The libraries the components vendor (corosio, the CRDT library, the full-text search library, the HTTP libraries) are pinned by those components.
Every component's images are on Docker Hub under oramasearch/, tagged with the version above.
What it includes besides the pins
- The release manifest,
release.yaml, checked against the pins byscripts/check-release-manifest.py. - The installer,
install/install.sh, and four install guides: Try it, Run it, Scale it and Air-gapped (docs/install/). - The licence and contributor files:
LICENSE(MIT),COPYRIGHT,THIRD_PARTY_LICENSES,SECURITY.md,CONTRIBUTING.md,CLA.md,CODE_OF_CONDUCT.mdandTRADEMARKS.md. - Governance:
GOVERNANCE.md(Orama-led, with maintainers for each component, and RFCs for larger decisions),MAINTAINERS.mdandROADMAP.md. - The agent guide (
CLAUDE.md) and thelakeskill, so an agent can work with the platform:claude, then/lake.
What it does not include
- The one-command
tryimage (lake-try). Until it ships, Try it runs the stack on a single-node k0s or k3s: directly on Linux, or in Docker on macOS and Windows. - Deferred work in the components: each component's release notes list what it does not include and the pull requests deferred to a later release.
Known issues
- Some images bundle GPL or LGPL userland from their base images (Debian or Alpine), and a few binaries link LGPL libraries. Each component's notes name its own. They are being moved to static bases in later releases.
- k0s's embedded liquid build was not run to the end with liquid v0.3.0; the clone of it is verified, the binary is not.
- The manifest's image digests are empty, and
publishedisfalse, until the images are signed.lake-mcp-runneris not pinned here. - Each component's known issues are on its own release page, linked above.
Verification
| Check | Result |
|---|---|
| Every pin is a released version | Each component's pinned commit is the commit of its latest release tag, as listed above (k0s, a fork, by commit) |
scripts/check-release-manifest.py |
passes: every commit in release.yaml is the commit this repository pins |
| Anonymous recursive clone | git clone --recurse-submodules with no SSH key and no credentials: 67 submodules, all fetched over HTTPS, exit 0 |
tools/spdx-headers.py --check |
passes |
This repository has no build or test suite of its own; each component's release page has its counts.