Billion random sources,
one lens.

A vector database stores embeddings — it doesn't know which sources are stale, contradictory, or out of scope for the requesting user, and it certainly can't deploy software. LensHub adds health scoring, access control, deduplication, feedback loops, versioning, and an app deployment path on top of retrieval. You get production-grade context, not just search.

When an agent builds something usable — a dashboard, a small internal app, a report — it can deploy it through LensHub with a single tool call. Colleagues open it with their normal company login, and it runs on your own infrastructure. Each app is isolated by default with no access to anything outside itself; every connection it needs is requested by the agent and approved by an admin, who sees exactly what is being asked for. Start, stop, logs, status, deletion and a network map are all in the UI.

Self-hosted: never. Everything runs in your VPC — your data never leaves, including anything your agents build and deploy. Managed: data is processed in your dedicated tenant. We will never train on your data.

Access is enforced at four layers. (1) Credential-scoped ingestion: a connector can only see what its token can see — a Slack bot invited to specific channels, a Confluence token scoped to certain spaces, a read-only GitHub key. Content the credential can't reach never enters the index at all. (2) Private by default: personal connectors are visible only to their owner. No role can read another user's private content — not even admins. (3) Explicit sharing: share a connector with specific teammates at read, write, or manage level. Grants take effect in seconds — and revoke just as fast. (4) Roles and audit: four org roles across 22 distinct permissions gate every action, and every administrative change lands in an append-only audit log. Visibility is enforced inside each query rather than filtered afterwards — even discovering that a restricted repository exists respects permissions.

It is told so. Every system that indexes company content skips some of it — files too large, excluded folders, vendored dependencies, configuration rules. LensHub keeps a per-repository record of what was skipped and why, and exposes indexing health and freshness to agents before they search. That way “I couldn’t find it” never gets mistaken for “it doesn’t exist.”

Any agent that supports the Model Context Protocol (MCP) — Claude, Claude Code, Cursor, Windsurf, or your own background workers — connects with a single command and gets all 24 tools, scoped to the person asking.

Self-hosted means your data never leaves your VPC. Managed runs in your chosen region (US or EU), encrypted at rest with AES-256 and in transit with TLS 1.3, with separate encryption keys per tenant. On managed, no LensHub employee has standing access — production access requires a JIT approval flow with two-person sign-off, fully audited. You can export everything at any time: every context, version, memory and audit record in a documented format. After cancellation all customer data is purged within 30 days with a deletion certificate. No exit fees, no proprietary formats, no lock-in.

The self-hosted version deploys in about 20 minutes. Managed onboarding takes a short call to configure your connectors and access scopes. Most teams have their first agent query working within an hour. Setup, connectors, permissions and self-hosting are all documented at docs.lenshub.ai.