What we're building

Teams shipping AI on private knowledge keep hitting the same wall: generic RAG returns fragments, agents sound confident when they’re not, and standing up ingest, structure, retrieval, and governance takes quarters. Restormel exists so you can own the knowledge your agents rely on — reviewable, keyed to your providers, and wired into the same control plane as routing and quality.

Our focus now is Connect: turn documents into a graph you trust, serve it to agents, and check claims before they reach users. Founders Circle members help set what ships next.

Already live

On restormel.dev today — see the changelog for detail.

  • Your knowledge graph, under your control

    Connect your sources, choose your graph store, and route every AI step with your own keys — no separate RAG platform to stand up.

  • Ingest with a human in the loop

    Documents become structured ideas you can review, flag, and re-check — so agents build on what you’ve approved, not a black-box chunk index.

  • Context agents can call

    Retrieve and verify from your workspace so copilots, apps, and MCP workflows can ground answers in what you actually know.

Building next

Where we see the most user value — reliable grounding, operator trust, and faster time-to-first graph.

  • Agents that actually use your knowledge

    Production-grade retrieval from your graph — the right ideas and relationships at query time, not empty or degraded context when it matters.

  • Trust you can ship on

    Validation and review at scale: clear supported vs weak claims, source provenance, and workflows to clear backlogs before answers reach users.

  • First trustworthy graph in an afternoon

    Onboarding for teams without a platform group — from first documents to a graph you’d let an agent cite, with fewer dead ends and surprises.

  • Embed without rebuilding the pipeline

    Stable APIs and docs for product teams who need ingest, retrieve, and verify in their app — not a fork of someone else’s notebook.

On the horizon

Committed direction after the core Connect loop lands — shaped by customer signal, not a feature laundry list.

  • Catch knowledge regressions before users do

    Tie Connect to Testing so model, route, or corpus changes get the same acceptance scrutiny as the rest of your AI stack.

  • See cost and quality in one place

    Spend and reliability across ingest, retrieval, and routing — so you can tune budgets and spot when grounding quietly degrades.

  • Start from how your field thinks

    Curated domain starting points (research, ops, policy, and more) so structure matches expert mental models, not generic “chunks.”

  • Enterprise-ready trust

    Audit exports, policy bundles, and rollout patterns for teams that must show where an answer came from — exploring demand with Founders partners.

Community

Suggest a feature

Open a discussion on GitHub — we read every thread and fold signal into planning.

GitHub Discussions →

Founders Circle members get a vote

Early partners help set priorities with direct access to what ships next — not just a discount.

Learn about Founders Circle →

Timelines are indicative, not commitments. Shipped work is in the changelog; technical programme notes live in the GitHub repository.