Kelvar Get started

Autonomous cyber defense

Kelvar's always-on agents watch your code, cloud configuration, endpoints and alerts, forge the fix, and record every action in a ledger you can verify.

Get started How it works

macOS 14+ on Apple Silicon. Research preview.

Connects to GitHub today. Cloud guards are in preview

Findings you can act on, not alerts to triage.

Connect a repo and Kelvar audits it, keeps only what it can confirm, and returns a verdict you can gate a merge on.

target-appsample repo, planted flaws
webhook_handler.py:4

Unsafe deserialization sink. Zero-click RCE vector.

critical
requirements.txt:1

requests 2.30.0 is affected by CVE-2023-32681.

medium
Verdict
ScanningBlock
  • 1 critical
  • 1 medium

Merge held until it is fixed.

Real output from the built-in scanner on a small sample repo with planted flaws. The agents go deeper than this first pass.

One audit, end to end.

Ask an agent to audit a folder it can't see yet. It builds itself a read-only room, reads, and reports. Nothing is trusted until a check passes.

  1. chat

    You ask

    "Audit target-app for zero-click and CVE risk." The Zero-Click Rapid Defender picks it up.

  2. authoring.apply

    It builds a read-only room

    A child worker is authored with read access to that one path and nothing else.

  3. file.read

    It reads the code

    Source files and dependency manifests, inside a sandbox limited to the admitted path.

  4. security.finding

    Findings are recorded

    Six confirmed on the sample repo: three critical, two high, one medium. Each carries file, line and advisory, and lands on the Findings screen.

  5. file.write

    Hand one to the patch agent

    It prepares a minimal fix in an isolated git worktree. A deterministic verifier must pass before the ledger records a mark.

Seven agents. Every side covered.

Each agent has one job, a sandboxed room and a policy, so every check runs with only the access it needs.

Zero-click and CVE response

Traces untrusted input to dangerous sinks, then drafts a virtual patch or a narrower boundary before attackers find the way in.

Cloud and server config

Live read-only scanning of Linux servers over SSH tunnels with a hardened collector. Reviews Terraform, CloudFormation, Kubernetes and Dockerfiles for AWS, Google Cloud and Azure today; live cloud scanning is next.

Code and supply chain

Audits lockfiles, workflows and look-alike packages on every pull request.

Secrets and data leaks

Finds leaked API keys, cloud credentials and private keys in code, git history and CI logs, and maps what each exposed key can reach.

Endpoint and browser

Opens suspicious links and OAuth consent pages in an isolated browser, and audits the device it runs on for startup items, cron jobs, extensions and open sockets. A small background service will extend this to every laptop in your fleet.

Fleet monitoring coming soon

Patch and remediation

Turns a confirmed finding into a minimal fix in an isolated worktree, runs the checks, and keeps a reversible change journal.

24/7 SOC

Investigates every SIEM, identity and webhook alert, links signals across surfaces, escalates only incidents it can confirm, and keeps SOC 2, PCI and ISO 27001 evidence tied to ledger entries.

One watch across your code, cloud and endpoints.

Kelvar defends what you build, the infrastructure it runs on and the machines that touch it, and it joins up with the tools you already run.

What Kelvar covers

Code and dependencies, cloud and server configuration, leaked secrets, zero-click sinks, suspicious links and endpoint persistence, plus the alerts from your SIEM. Live cloud scanning and fleet-wide laptop monitoring are coming next.

Meet the agents

What it joins up with

Alerts from your endpoint, identity and SIEM tools arrive over webhooks. An agent investigates each one, links it to related signals across your projects, and escalates only what it can confirm.

See the webhook API

How it is different

Priced per completed run instead of per seat. Runs on your machine or appliance. Works with any model, including local ones. And every action sits in a ledger you can verify yourself.

Read the security model

Connect once. Kelvar handles the rest.

  1. Connect

    Add GitHub or a SIEM feed. Cloud guards are in preview. Access stays read-only until you opt in to write-back.

  2. Watch

    Agents audit continuously and record each confirmed finding with the evidence behind it.

  3. Block

    Risky merges and payloads are stopped at the gate. Fixes arrive as small patches built in an isolated worktree.

Every action leaves proof.

Findings, blocks and patches are written to an append-only, hash-chained ledger. A result is marked verified only after a deterministic check passes, never on a model's say-so.

How the ledger works
    Illustration. Each hash is computed in your browser from the entry before it.

    Free for open source. Pay per run for everything else.

    No seats. You pay for the runs your agents complete, and every charge maps to a run in your ledger.

    Community

    $0 for open source projects

    Every agent, the full ledger and the Findings screen on public open source repositories. Bring your own model keys, including local models, and pay your provider directly.

    Run it free

    Usage

    from $0.02 per run, no seats

    For private repositories, companies and any other use. Pay per completed run, with your own keys or ours. Set a monthly cap and agents stop when they reach it.

    See usage pricing

    Enterprise

    Committed on-premise from $75k per year

    The appliance inside your own network, volume run pricing, and a signed audit trail for SOC 2. Or let the Kelvar team run the agents for you, from $5k per month.

    See Enterprise

    What a run costs

    Scan run
    $0.02
    Deterministic checks for secrets, dependencies and infrastructure code. No model involved.
    Agent run
    $0.40
    A model-reasoned audit, triage or patch at standard effort, in a sandboxed room.
    Deep run
    $1.50
    High-effort audit that builds child workers and verifies every finding before it is recorded.
    • Your keys or ours. With your own keys you pay your provider directly and Kelvar bills the run only. With ours, model cost passes through at cost plus 15%.
    • Only completed runs. Runs that fail, error out or are refused by policy are never billed.
    • Fail-closed cap. Set a monthly ceiling. When it is reached, agents stop instead of overspending.

    Questions security teams ask.

    Does Kelvar change my code or infrastructure?

    Not unless you allow it. Guards connect read-only, write-back is opt-in, and every change an agent makes is recorded in the ledger with the evidence that justified it.

    Where does it run?

    On your own machine or on an on-premise appliance. Scans are local-first, and nothing leaves your environment except the model calls you choose to enable.

    Which models can it use?

    Any provider you connect, including local and custom models through an OpenAI-compatible endpoint. The kernel does not depend on a specific vendor.

    Does it replace endpoint protection or cloud scanners?

    Not today. Kelvar reviews cloud configuration in code and checks the device it runs on for persistence. Live scanning of AWS, Google Cloud and Azure accounts, and a small background service that monitors laptops across a fleet, are in development. Until they ship, keep your current tooling and send its alerts to Kelvar for triage.

    What happens when it finds something?

    Confirmed findings appear on the Findings screen with file, line and advisory. From there you can hand one to the patch agent, which prepares a minimal fix in an isolated worktree.

    Put a watch on everything you ship.

    Get started