-
Notifications
You must be signed in to change notification settings - Fork 4
Roadmap
antonyrag edited this page Aug 17, 2026
·
11 revisions
Full detail lives in
ROADMAP.md
at the repo root. This page summarizes current status.
| # | Item | Status |
|---|---|---|
| 1 | SECURITY.md |
✅ Done |
| 2 | Dependency SBOM (auto-regenerated in CI) | ✅ Done |
| 3 | CodeQL + Dependabot | ✅ Done — already caught a real CVE in a dependency |
| 4 | Real, measured benchmarks | ✅ Done — see Benchmarks |
| 5 | Migration guide | ✅ Done — see Migration Guide |
| 6 | Live-verify Pinecone/Weaviate/Qdrant/Milvus | 🔒 Blocked — no real accounts available yet |
| 7 | Live-verify Mistral/Together/Cohere/Voyage embeddings | 🔒 Blocked — no real accounts available yet |
Items #6 and #7 are blocked purely on account access, not remaining engineering work — the code is written and tested against each SDK's real source, just not run against a live service yet.
| Order | Package | Purpose | Status |
|---|---|---|---|
| 1 | ragleap-rag |
RAG engine core | ✅ v0.12.2, hardening phase above |
| 2 | ragleap-graph |
Neo4j knowledge-graph retrieval | ✅ v0.6.5 — LLM entity/relation extraction, hybrid vector+graph retrieval, per-document idempotency, per-document lineage lookup (find_lineage), per-user data isolation (user_id=), full CHANGELOG at packages/ragleap-graph/CHANGELOG.md |
| 3 | ragleap-integrations |
MCP + WhatsApp/Telegram/Discord/Voice | Not started |
| 4 |
ragleap-agents + ragleap-flows
|
Role-based crews, low-code orchestration | Not started |
| 5 | ragleap-observability |
Tracing + LLM eval framework | Not started, business decision pending |
| 6 | ragleap-ops |
Docker/K8s/Helm/CI-CD templates | Ongoing, grows alongside everything |
| 7 |
ragleap (meta-package) |
pip install ragleap = everything bundled |
Not started |
More than once, an externally-generated planning document proposed expanding into a 14-library "AGI ecosystem" with 3 libraries framed as "world-first monopoly" products, plus specific benchmark numbers to publish directly.
This was reviewed twice and rejected both times:
- The proposed benchmark numbers were fabricated — no such measurements were ever run. Publishing them would have directly contradicted this project's core standard.
-
The "3 monopoly libraries" already exist in
ragleap-rag, just renamed — fallback chaining (fallbacks=, v0.6.4) and guaranteed structured output (response_format=, v0.8.0) aren't new, unclaimed categories. - It was scope creep away from what's actually working.
Full reasoning: see the "Considered and Rejected" section in ROADMAP.md.