Looking for the self-hosted AI employee system? This site documents the open-source,
pip install-able packages. To install and run ragleap-core (AI employees, channels, Docker), see docs.ragleap.com; the project home is ragleap.com.
11 packages shipped (10 Python + 1 Java) · 741 Python test functions · every count below cross-checked against real code
RagLeap Packages
Open, honestly documented Python libraries for retrieval-augmented generation and knowledge-graph retrieval. No projected numbers — every claim on this site is either live-verified or labeled as not yet verified.
10
Python packages on PyPI
741
Python test functions (300 + 112 + 129 + 73 + 94 + 29 + 4)
6
Vector backends
8
Embedding providers
Packages
All three are part of the same public monorepo — ragleap-graph composes with ragleap-rag for hybrid vector+graph retrieval, and ragleap-vectorstores extends ragleap-rag's VectorBackend interface with additional backends — but each installs and works independently.
v0.13.0
ragleap-rag
A narrow, opinionated hybrid retrieval engine. Six vector backends, eight embedding providers, twelve-plus generation providers, twenty-eight ingestion formats.
v0.12.0
ragleap-rag (Java)
Java port of the ragleap-rag core pipeline, on Maven Central. Connection pooling, embeddings, all 6 vector backends (pgvector, FAISS, Qdrant, Weaviate, Milvus, Pinecone), retrieval, generation (streaming, structured output, vision), conversation memory, query rewriting, and a 27-format document parser.
v0.9.0
ragleap-graph
Knowledge-graph retrieval over Neo4j. LLM entity extraction, deduplication, typed relations, and hybrid vector+graph search.
v0.5.1
ragleap-vectorstores
Pluggable vector backends beyond ragleap-rag core's six. Chroma and LanceDB (embedded/local), Redis via RediSearch/Redis Stack, Upstash Vector (managed serverless), plus OpenSearch (k-NN, requires a running instance).
v0.4.1
ragleap-tools
Standalone, dependency-light tools for LLM tool-calling — OpenAI/Gemini-style function-calling schemas. Calculator, sandboxed file ops, date/time, unit conversion, JSON/CSV parsing, text utilities, document search, pluggable BYOK web search (Tavily, Serper), GitHub repository search, BYOK image description (Gemini, Anthropic), plus optional ragleap-rag-backed ingestion.
v0.1.0
ragleap-integrations
MCP client over Streamable HTTP that returns ragleap-tools Tool objects — owner-configured servers and an exact tool allowlist, modern and legacy protocol paths, and a standard-library-only transport with a pinned IP, verified TLS and a wall-clock deadline.
v0.1.0
ragleap-agents
Single-agent act-observe loop over ragleap-tools Tool objects. Schema-checked actions, untrusted tool results, and a taint rule that forces approval for outbound tools after untrusted content. Scripted-model tests only; not verified with a real model.
v0.5.0
ragleap-ops
Kubernetes Deployment and Service manifests, Helm chart, NetworkPolicies, Ingress+TLS, multi-environment values, and Backup/DR for RagLeap Core — all live-tested on a real kind cluster.
v0.3.0
ragleap-app-chart
Generic, reusable Helm chart for deploying arbitrary services to Kubernetes — not RagLeap-specific. Point it at your own app via a services: list and range loops.
v0.1.0
ragleap-terraform
Terraform module that creates a local kind cluster and installs the RagLeap Helm charts. Applied and destroyed on a real cluster; no cloud modules yet.
v0.6.2
ragleap-observability
Prometheus, Grafana, Loki + Promtail for ragleap-ops, live-verified end-to-end against a real cluster — real connection confirmed, real non-zero metrics returned, 21+ real log streams with correct namespace/pod/container labels; AlertManager wired to Prometheus with a first real alert rule and optional Slack/email receivers (off by default).
Install
All seven are on PyPI and installable independently.
ragleap-rag
pip install ragleap-rag # or, with uv uv add ragleap-rag
ragleap-rag (Java)
<dependency> <groupId>io.github.antonyrag</groupId> <artifactId>ragleap-rag</artifactId> <version>0.12.0</version> </dependency>
ragleap-graph
pip install ragleap-graph # or, with uv uv add ragleap-graph
ragleap-vectorstores
pip install ragleap-vectorstores[chroma] # or, with uv uv add ragleap-vectorstores[chroma] pip install ragleap-vectorstores[lancedb] # or, with uv uv add ragleap-vectorstores[lancedb] pip install ragleap-vectorstores[redis] # or, with uv uv add ragleap-vectorstores[redis] pip install ragleap-vectorstores[upstash] # or, with uv uv add ragleap-vectorstores[upstash] pip install ragleap-vectorstores[opensearch] # or, with uv uv add ragleap-vectorstores[opensearch]
ragleap-tools
pip install ragleap-tools # or, with uv uv add ragleap-tools pip install ragleap-tools[ingest] # optional, needs ragleap-rag
ragleap-ops
pip install ragleap-ops # or, with uv uv add ragleap-ops
ragleap-app-chart
pip install ragleap-app-chart # or, with uv uv add ragleap-app-chart
ragleap-terraform
pip install ragleap-terraform # or, with uv uv add ragleap-terraform
ragleap-observability
pip install ragleap-observability # or, with uv uv add ragleap-observability
Ecosystem roadmap
The full plan, in build order. Four shipped, nine more to go — statuses reflect real progress, not intent.
| Order | Package | Purpose | Status |
|---|---|---|---|
| 1 | ragleap-rag | RAG engine core | ✓ v0.13.0 |
| -- | ragleap-rag (Java) | Java port of the ragleap-rag core pipeline, on Maven Central (not part of the numbered sequence, independently versioned from the Python package) | ✓ v0.11.0 |
| 2 | ragleap-graph | Neo4j knowledge-graph retrieval | ✓ v0.9.0 |
| 3 | ragleap-vectorstores | Pluggable backends beyond core defaults | ✓ v0.5.1 |
| 4 | ragleap-tools | Stateless tools + sandboxed file ops + document search + web search + image description for LLM tool-calling (code-exec/HTTP-fetch/connectors deliberately deferred) | ✓ v0.4.1 |
| 5 | ragleap-integrations | MCP client over Streamable HTTP that returns ragleap-tools Tool objects (curated native connectors and channel integrations not built yet) | ✓ v0.1.0 |
| 6 | ragleap-agents | Single-agent act-observe loop over ragleap-tools with approval gates (multi-agent crews planned) | ✓ v0.1.0 (scripted model only; not verified with a real model) |
| 7 | ragleap-flows | Low-code orchestration + HITL, bootstrapped from existing n8n integration | not started |
| 8 | ragleap-observability | Tracing, hallucination detection, LLM-as-judge evaluation framework | not started |
| 8⚠️ | (naming note) | The PyPI name ragleap-observability is already taken by a separate, shipped package below (infra monitoring). This roadmap item (LLM tracing/eval) needs a different package name when built. | n/a |
| -- | ragleap-observability | Prometheus, Grafana, Loki + Promtail, and AlertManager for ragleap-ops, live-verified end-to-end (not part of the numbered sequence) | ✓ v0.6.2 (live-tested on a real kind cluster) |
| 9 | ragleap-ops | Docker/K8s/Helm/CI-CD templates, IaC | ✓ v0.5.0 (manifests, Helm chart, NetworkPolicies, Ingress+TLS, multi-env, Backup/DR) |
| -- | ragleap-app-chart | Generic, reusable Helm chart for arbitrary services (not RagLeap-specific, not part of the numbered sequence) | ✓ v0.3.0 (live-tested on a real kind cluster) |
| -- | ragleap-terraform | Terraform module: local kind cluster + RagLeap charts (not part of the numbered sequence) | ✓ v0.1.0 (applied and destroyed on a real kind cluster) |
| 10 | ragleap-studio | Visual low-code builder UI | not started |
| 11 | ragleap-memory | Long-term memory & context management | not started |
| 12 | ragleap-intelligence | Autonomous multi-step execution with optional human-in-the-loop approval | not started |
| 13 | ragleap | Meta-package — everything bundled | not started |
| 14 | ragleap-physical | Evolution: Embodied AI — MQTT/ROS/PLC bridge, sensors → actions | Evolution — after 1-13 |
Correction: an earlier version of this page said ragleap-vectorstores was folded into ragleap-rag's core rather than shipped separately — that plan changed. It shipped as its own package at v0.1.0, with Chroma as its first backend; see its reference page.
The standard
Every claim in code, docs, and this site is either live-verified against real infrastructure, or explicitly labeled unverified. When a past claim turns out wrong, a new entry documents the correction — nothing gets silently rewritten. See each package's reference page for the specific findings this held up against.