SVAROG is an AI coding terminal that thinks in plans, executes in a sandbox, and runs on infrastructure you control - including fully air-gapped. The agent power developers want, with the trust boundary enterprises require.
svarog> add a contact form with server-side validation ▸ PLAN (review before anything runs) 1. scaffold /components/ContactForm 2. add POST /api/contact with input validation 3. wire client fetch + error states 4. add a sandbox test, run it, report results svarog> proceed with the plan ✓ files written · ✓ test run in sandbox (no network) · ✓ 4/4 passing # ! to run shell · ↑↓ history · /plan to re-enable plan-first
The current generation of agentic coders is powerful but opaque: they act first and explain later, run with broad access, and ship your source to someone else's cloud. For regulated, security-conscious, and sovereignty-bound teams, that is a non-starter.
Autonomous agents edit, install and execute with little visibility. One wrong assumption becomes dozens of unwanted changes.
Prompts and repositories are streamed to third-party model APIs - unworkable under strict IP, data-residency or air-gap rules.
You are tied to one vendor's model, pricing and roadmap - with no way to route to a local model or self-host the brains.
SVAROG is built around three principles that flip the defaults: review intent before execution, keep work inside a controlled sandbox, and let the operator choose the model - including a fully local one.
{
"status": "ok",
"version": "2.1.0",
"providers": {
"svarog": true, # self-hosted, default
"anthropic": false, # opt-in BYO key
"gemini": false # opt-in BYO key
}
}
SVAROG is a single terminal interface backed by a self-hostable engine. Natural language becomes a reviewable plan; shell and code execution happen inside a sandbox; results come back inline.
Type what you want in plain language. Use ! to drop into a sandboxed shell, /plan to re-enable plan-first, and ↑↓ to walk history.
SVAROG returns a structured, numbered plan. Approve it ("proceed with the plan"), revise it, or stop - nothing executes until you say so.
Code runs in an isolated sandbox with a no-network bash mode. The model proposes; the sandbox contains the blast radius.
Run code, preview output, connect a GitHub repo, and download the result. Air-gap mode keeps the whole loop offline.
Run code and shell commands in isolation - output inline, network off when you need it off.
Connect a repository and let SVAROG work in context, with changes you reviewed first.
Every build starts as a reviewable plan - auditable intent before any mutation.
A no-network posture for classified, regulated or sovereignty-bound work.
AI-assisted development is becoming standard practice - but the fastest-growing tools route source code through third-party clouds. That leaves an entire segment of high-value buyers underserved.
Finance, healthcare, defense and public sector teams bound by data-residency and IP-confidentiality rules.
Platform and security orgs that want agentic productivity without an uncontrolled execution surface.
Founders and consultancies who simply do not want client code training someone else's model.
Market framing is qualitative and illustrative; no specific sizing, revenue or customer figures are claimed.
A developer-friendly free tier drives adoption; paid plans monetize the self-hosting, air-gap and governance capabilities that enterprises require. Illustrative structure shown below.
Pricing tiers are illustrative of the intended model and not a current commercial offer.
Plan-first execution and a contained sandbox make every action reviewable. Trust is the product surface - not a setting buried in a menu.
A self-hostable backend with a native model means the agent comes to the code, not the other way around. Air-gap is a supported mode, not a workaround.
The same terminal routes to a local SVAROG model or an opt-in frontier API. Customers optimize for cost, latency or capability - without re-tooling.
Chat, plan, shell, code execution, preview and GitHub live in a single terminal. Less context-switching, fewer integrations to govern.
SVAROG pairs a lightweight web terminal with a backend API (currently v2.1) that abstracts providers behind one interface - so the front end stays the same whether the brains are local or remote.
Source & integration reference: github.com/nattimmis/svarog-embedded-coder
AI coding agents are now in everyday workflows - and so are the questions about where code goes and what the agent is allowed to do.
Local and open-weight models are finally good enough to run real coding workloads inside a private boundary - making self-host practical, not theoretical.
Data-residency, IP-protection and sovereignty mandates are hardening. The market needs an agent that satisfies them by architecture.
SVAROG comes from a developer-tools team that builds for the boundary cases: regulated industries, sovereign deployments and security-conscious shops. The conviction is simple - agentic coding should be powerful and private, with a human approving the plan and infrastructure the customer controls.
Deep experience shipping fast, low-friction terminal and CLI experiences developers actually keep open.
Focused on isolation, controlled execution and the architecture air-gapped customers depend on.
Practical model routing and plan-first orchestration that puts a reviewable step between intent and action.
Try the live SVAROG terminal, connect a repository, or talk to us about a self-hosted and air-gapped deployment.