Kremis Breakdown
Verifiable knowledge substrate for AI agents
1 community upvotes on Launch Llama
Categories: Developer Tools Artificial Intelligence Open Source
Key facts
- Maker: TyKolt
- Categories: Developer Tools, Artificial Intelligence, Open Source
- Community upvotes: 1
- Website: https://github.com/TyKolt/kremis
About Kremis
Kremis is a verifiable knowledge substrate for AI agents: instead of generating answers and hoping they’re right, agents query a deterministic graph that returns honest output — facts, inferences, or an explicit unknown, never silent gap-filling — with every result traceable to a graph path and inspectable state hash; queries can also return a Verifiable Query Certificate, a deterministic reproducible object that can be re-run against the same graph state to re-verify the result, including certified unknown (proof of absence); shipped as an MCP server (queryable directly from Claude) and an HTTP API, open source (Apache 2.0), built in Rust, with in-memory or crash-safe persistent storage.
Who Is Kremis For?
For developers shipping LLM features who need knowledge they can trust
- AI agent developers
- Teams shipping LLM features
- Engineers who need auditable knowledge
- Claude / MCP users
How Founders Can Use Kremis
- Agent memory via MCP
- LLM fact-checking
- Provenance & audit trail
- Reproducible, re-verifiable query results (certificates)
Kremis vs Alternatives
- GitHub Copilot
- Linear
- Vercel
Kremis Features
Kremis is small on purpose. Every feature serves one goal: knowledge an AI agent can query without inventing.
- Honest output: facts, inferences, or "unknown"
- Deterministic: same input, same output
- Verifiable Query Certificates — reproducible proofs, incl. proof of absence
Pricing
Free and open source (Apache 2.0). Self-hosted — no account, no usage limits, no telemetry.
Verdict
Kremis is experimental, alpha-stage — not a turnkey product. For developers tired of confident hallucinations, it takes a different approach from vector-store guesswork: a deterministic graph that answers with a fact, an inference, or an honest "I don't know", tracing every result back to the underlying data. Results can also be packaged as a Verifiable Query Certificate that another party can reproduce to independently re-check the answer. Small, inspectable, and open source.
Frequently Asked Questions
What is Kremis?
Verifiable knowledge substrate for AI agents
Who is Kremis for?
For developers shipping LLM features who need knowledge they can trust • AI agent developers • Teams shipping LLM features • Engineers who need auditable knowledge • Claude / MCP users
How much does Kremis cost?
Free and open source (Apache 2.0). Self-hosted — no account, no usage limits, no telemetry.
What category does Kremis belong to?
Kremis is listed under Developer Tools, Artificial Intelligence, Open Source on Launch Llama.
How do I get started with Kremis?
Click the "Visit Website" button on this page to go directly to Kremis. You can also upvote and leave a review to help other founders discover it.
Is Kremis production-ready?
No — it's experimental and alpha-stage. Stable enough to build and test against, but the API may change.
What is a Verifiable Query Certificate?
A deterministic, reproducible record of a query result (introduced in v0.19.0). Given the same graph state, anyone can re-run it and get byte-identical output — even an "unknown" answer is certifiable as a proof of absence. Re-verification reproduces the canonical state, so it is not a standalone proof in isolation.
← Browse all products · Launch Llama home · Listed on Launch Llama at /products/kremis