Multi-provider LLM access
A first-party abstraction over 40+ providers — Anthropic/Claude, OpenAI, DeepSeek, Gemini, Mistral, Groq, xAI/Grok, and local Ollama as a fallback — behind one interface with retries, circuit breakers, and health checks.
// AI-native software engineering
AI-native software engineering, built to be trusted.
We build AI development systems — the models, agents, orchestration, and infrastructure that turn large language models into dependable software — together with the governance layer that keeps them honest.
40+
LLM providers, one interface
140+
repositories, one constitution
33
products across three tiers
// about
A fleet, not a monolith.
Vasic Digital is a focused engineering practice building an interconnected family of AI-development products and reusable modules. Rather than one monolith, the work is organized as a fleet: large product applications on top of dozens of small, independently-tested, decoupled modules — so proven parts are reused across every product instead of rebuilt.
The backbone language is Go, complemented by Kotlin & Kotlin Multiplatform, TypeScript/React, Python, Swift, and Shell — chosen per job. What ties the fleet together is discipline made mechanical: every project inherits a shared engineering Constitution as a Git submodule, and every capability a product advertises must be backed by an automated, evidence-producing test. A feature is not "done" when the tests pass; it is done when a real user can actually use it, and there is captured evidence to prove it.
// what we do
AI-based development, end to end.
We build the substrate for AI systems — the hard part beneath the app.
Agent orchestration
Headless CLI coding-agent control planes, graph-based agentic workflows, multi-round "AI debate" consensus, and DAG/pipeline runtimes.
LLM verification
A trust layer that scores models with a mandatory comprehension gate ("Do you see my code?") plus latency, streaming, function-calling, vision, and embeddings tests — exporting a verified-only configuration.
Retrieval & memory
RAG, vector databases, embeddings, and fused agent-memory engines (Mem0 + Cognee + Letta) with infinite-context compression.
Defensive LLM
Guardrails, PII detection, adversarial red-team fixtures, and input canonicalisation — safety built into the substrate, not bolted on.
// tier 1 — flagship
The Helix product family & LLM infrastructure
Our flagship line spans the full AI-development lifecycle. In priority order.
HelixTrack beta
GoGinHTTP/3 QUICPostgreSQL+8
A free-world alternative to JIRA — the flagship of the Helix-Track line.
HelixAgent beta
GoGinPostgreSQLLLMsVerifier+6
An ensemble LLM service that lets multiple models debate and ships the answer they agree on.
HelixCode beta
GoGinSSHModel Context Protocol+2
A distributed AI development platform that divides work across SSH-managed workers with checkpoint/rollback.
HelixLLM beta
GoHTTP/3 QUICllama.cppLLMsVerifier+5
One binary, six modes: OpenAI- and Anthropic-compatible inference from laptop to cluster, over HTTP/3.
HelixCluster in-development
GoZiggRPCRaft + SWIM+7
A distributed operating system for AI compute, from datacenter GPUs to edge handhelds.
LLMProvider beta
Gonet/httplogrusdigital.vasic.models+2
One interface, 43 providers — with circuit breakers, retries, and health baked in.
LLMOrchestrator beta
GoJSON-lines over stdioCircuit breakerHealthMonitor+2
One control plane for every headless CLI coding agent.
LLMsVerifier beta
GoGinSQLite + SQLCiphergRPC+6
Verify. Monitor. Optimize. The single source of truth for LLM and provider verification metadata.
HelixMemory beta
GoMem0CogneeLetta+3
One memory brain for AI agents — four best-in-class engines, fused.
// quality & governance — our differentiator
Governance as an auditable fact, not an aspiration.
Two pillars make the whole fleet coherent and trustworthy.
HelixConstitution shipped
A universal, project-agnostic engineering rulebook shipped as a Git submodule and inherited by every project across a 140+-repository fleet. It encodes non-negotiable discipline — anti-bluff evidence gates, false-positive immunity, data and host safety — that a project may extend but never weaken. One submodule bump upgrades the rules everywhere, and every gate is paired with a mutation test that proves the gate itself isn't a sham.
HelixQA beta
Anti-bluff QA orchestration. It runs written YAML test banks and fully-autonomous, LLM-plus-computer-vision QA sessions across Android, Android TV, Web, and Desktop — and it refuses to score a PASS without captured runtime evidence: screenshots, logcat, video, stack traces. "We tested it" becomes "here is the video, the logcat, and the ticket."
// tier 2 — utilities
Vasic Digital utilities
Product-grade tools that stand on their own.
Catalogizer production
GoGinReactWebSockets+10
Advanced multi-protocol, encrypted media collection management — detect, catalog, and enrich everything you own.
Courses-Creator production
GoReactElectronReact Native+8
Markdown in, professional video course out — AI-enhanced, multi-platform.
VisionEngine active
GoGoCV / OpenCVLLM visionGraph algorithms
See the UI like a user — computer vision plus LLM vision for analysis and navigation.
DocProcessor active
GoLLM agentsHeuristic parseri18n
Turn documentation into a verifiable feature map for QA automation.
Docs Chain active
GoDAGSQLitefsnotify
No tracked document can fall out of sync — content-hashed, bidirectional, atomic.
Herald active
GoShellLLM intent inferenceDocs Chain
Every alert reaches the right destination — no command syntax required.
task_bridge scaffold
GoSQLiteHMAC-SHA256cron + webhooks
Your task board and your source-of-truth, impeccably in sync — both ways.
Reusable Module Suite scaffold
Go (digital.vasic.*)Kotlin MultiplatformVector DB / RAG / MCP+1
Build once, reuse everywhere — a fleet of small, decoupled, independently-tested Go and KMP modules.
// tier 3 — heritage
Server Factory — infrastructure automation
Our DevOps heritage: declarative JSON turned into fully-provisioned, Dockerized servers.
Mail Server Factory production
KotlinShellDockerQEMU+4
Run your mail server like the boss — describe it in JSON, deploy it anywhere.
Server Factory Core Framework stable
KotlinShellGradle
The shared engine behind every Server Factory.
Qemu-Utils active
ShellQEMUBridge / TAPISO images
QEMU VM images, managed like artifacts — download, run, network, publish.
Parallels-Utils active
ShellPython 3Parallels
Compress, publish, and reuse your Parallels VM images across every machine.
// technologies
Grounded in our actual stack.
Languages
GoKotlin / KMPTypeScriptPythonSwiftShellPL/pgSQLTLA+
AI / LLM
43+ provider adaptersMCPRAGVector DBsHiPlan / MCTS / ToTSWE-bench / HumanEval / MMLU
Backend
GingRPC + ProtobufHTTP/3 (QUIC)WebSocketsAngular / ReactKafka / RabbitMQ
Data
PostgreSQLSQLiteSQLCipherRedisNeo4jClickHouseMinIO / S3
Infra / DevOps
Docker & ComposeKubernetes + HelmPrometheus + GrafanaOpenTelemetryQEMU / LibvirtGitHub Actions
Testing / QA
HelixQA (anti-bluff)Challenge harnessesgo test -raceSonarQubesemgrep / gosec / trivy / snyk
// portfolio
One fleet, one discipline — see all 33 products.
A unified, evidence-based portfolio of the Helix family, the vasic-digital utilities, and the Server Factory toolchain.
// contact
Let's build something verifiable.
Anyone can wire an app to an LLM. We build the part that is hard: AI systems that are verifiable, reusable, and honest. We don't ask you to trust the green checkmark — we show you the evidence behind it.
Email
i@mvasic.ru
GitHub
github.com/vasic-digital