kvendra
·MENU

·SECTION 07 — COMMUNITY

Community.

Open-core community. Lab notes, frequently asked questions, the list of open issues we are tracking, and the source on GitHub. Every entry is anchored to a Kvendra KB entity so you can chase the original record yourself.

§ 01 — CHANNELS ·WHERE WE TALK
·GITHUB
·OPEN

GitHub Discussions

Source repos under github.com/KvendraAI: CLI, web, skills, platform, helm, reference stack. File issues and open RFCs here.

·X / TWITTER
·NEWS

@kvendraai

Release announcements + occasional lab notes. x.com/kvendraai →

·EMAIL
·LOW-VOLUME

community@kvendra.ai

For maintainers and partners — anything that doesn't fit on a public issue. community@kvendra.ai

§ 02 — ARTICLES & LAB NOTES ·HANDS-ON
·2026-05-22
·LAB

Free tier Kvendra — lab notes from the 2026-05 spike

End-to-end self-hosted Kvendra on a g4dn.xlarge with Tesla T4, Ollama 0.24, mistral-small3.1:24b, and Claude Code 2.1.147. What worked, what didn't, and why cline + local LLM is not production-ready yet.

READ ON GITHUB →

·SOURCE: PAT-KVD-819856 / ISSUE-KVD-SKILLS-BE4A0F

§ 03 — FAQ ·MOST-ASKED
    What problem does Kvendra solve?

    AI assistants forget. They lose context between sessions, they rewrite code that already exists, and they leak secrets into LLM context windows. Kvendra is a typed knowledge graph plus a local broker that survives the swap from one AI assistant to another. Your AI writes the code; Kvendra stops it from breaking the system.

    ·SOURCE: PAT-KVD-2CBB6D / DOC-KVD-A2AC97

    Is Kvendra free?

    The CLI is Apache-2.0 on crates.io. The Platform engine is AGPL-3.0 on GitHub. The community skills are Apache-2.0. You can run the whole stack self-hosted forever without paying anyone. You only pay when you want the hosted Enterprise KB with vault backup, multi- device sync, RBAC, and shared workspaces.

    ·SOURCE: ADR-KVD-1823F7

    What is the security model?

    Zero-knowledge vault at ~/.kvendra/. The master password never leaves the CLI process and never enters environment variables. A local broker mediates every sensitive primitive. Allowlists are HMAC-signed; manipulation is detectable. Audit log is a SQLite chain (HMAC chain kvendra/audit-hmac/v1).

    ·SOURCE: PAT-KVD-819856

    Which AI assistants can use Kvendra?

    Any MCP-native client. Claude Code today (with the optional skills plugin). Cursor, Cline, Continue, and any future MCP client consume the same 14 KB tools and 7 broker primitives directly. Your choice of assistant does not decide your governance model.

    ·SOURCE: DOC-KVD-A2AC97

    What is in the box today vs deferred?

    Today: typed graph entities, transactions, semantic search (BYOK embeddings or hosted), the 7 broker primitives, the 18-skill Claude Code plugin, multi-tenant SaaS with schema-per-tenant. Not today: code-signed binaries (deferred to v0.3.0+), production Helm chart (M3), cline + local LLM agent orchestration.

    ·SOURCE: ROAD-KVD-716183 / ISSUE-KVD-SKILLS-BE4A0F

    Can I bring my own embeddings provider?

    Yes. Self-hosted Platform exposes EMBEDDINGS_BASE_URL, EMBEDDINGS_MODEL, and EMBEDDINGS_API_KEY. Any OpenAI-compatible endpoint works — Ollama, OpenAI, Together, or Kvendra Cloud (200k free tokens/month).

    ·SOURCE: ADR-KVD-PLATFORM-562CE8

    What does the skills plugin do that the MCP tools don't?

    The MCP tools are the primitive layer. Skills are the orchestrator: a /new-feature invocation reads a SKILL.md plan, opens a transaction, dispatches sub-skills for requirements/plan/implement/test/validate, then activates or cancels the txn.

    ·SOURCE: ADR-KVD-SKILLS-552A8F

    How do I report a bug?

    File an issue on the relevant repo under github.com/KvendraAI. Critical or security-sensitive reports: email hello@kvendra.ai. Known issues are listed below.

§ 04 — KNOWN ISSUES ·LIVE STATUS

Open issues curated from the KB. Severity is medium or lower — anything higher is fixed in the next release window. Status is fetched at build time from GitHub when a public mapping exists.

  1. ·ISSUE-KVD-SKILLS-BE4A0F

    cline + local LLM tool-format fragility

    ·unknown ·SEV medium

    Use Claude Code + Ollama path instead (lab-validated end-to-end).

  2. ·ISSUE-KVD-ENTERPRISE-E52B74

    /v1/me JWT field inconsistency

    ·unknown ·SEV low

  3. ·ISSUE-KVD-ENTERPRISE-1A023D

    entity_update generate_embedding edge case

    ·unknown ·SEV low

  4. ·ISSUE-KVD-ENTERPRISE-7DE241

    component_id naming inconsistency (requires bare uppercase)

    ·unknown ·SEV low

  5. ·ISSUE-KVD-ENTERPRISE-DCB110

    ADR archive forbidden even in draft

    ·unknown ·SEV low

  6. ·ISSUE-KVD-ENTERPRISE-6CA184

    tenant_id display in dashboard (guiones vs underscores)

    ·unknown ·SEV low

  7. ·ISSUE-KVD-ENTERPRISE-DE7CCC

    entity_search default min_score filters valid 0.5+ matches

    ·unknown ·SEV low

  8. ·AWS-CASE-177850260300533

    AWS Bedrock InvokeModel quota pending

    ·unknown ·SEV medium

    Embeddings work via temporary cross-account proxy (ADR-KVD-030), retires 2026-07-19.

§ 05 — CONTRIBUTING ·OPEN-CORE