A Platform for Agents & Humans in The Loop
https://docs.vski.ai
- TypeScript 81.9%
- Python 10.2%
- Makefile 3.1%
- Go 2.4%
- JavaScript 2.3%
- Other 0.1%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
| .bin | ||
| .release | ||
| benchmarks | ||
| examples/plugins | ||
| integration | ||
| k8s@357ebc9104 | ||
| platform.wiki@4c5b53599e | ||
| py-sdk@6058e10191 | ||
| tango@0209dc2754 | ||
| ts-sdk@04ba5f97ca | ||
| vski@ea7864a5a5 | ||
| web@dac7f10460 | ||
| .env.example | ||
| .gitignore | ||
| .gitmodules | ||
| LICENSE | ||
| README.md | ||
VSKI
Vector-Semantic Knowledge Index
VSKI is a high-performance, lightweight backend engine designed for modern applications requiring real-time synchronization and complex distributed workflows. It provides a multi-tenant environment using SQLite3 with a focus on simplicity, scalability, and security.
VSKI can easily replace PostgreSQL or other traditional RDBMS for most backend-related tasks. It provides a server layer on top of SQLite3 ensuring access control, proper concurrent access, ACID, and high-throughput database I/O.
Features
- Multi-Tenant SQLite: automatic management of multiple isolated database files
- Dynamic Schema: flexible collection-based data management with runtime schema enforcement and reviewable migrations
- Declarative IaC: collections, pipelines, cron, webhooks, secrets, plugin
instances, graphs and connectors as versioned YAML (
vski plan/apply/pull), plus idempotent seed files (vski seed) - Distributed Workflows: persistent, stateful workflow orchestration with WebSocket-based workers
- Pipelines: declarative JSON DAGs (plugin calls, HTTP requests, timers, approval gates, LLM steps) executed durably by the engine
- Light Graphs: labeled nodes and weighted edges with in-memory traversals in the same per-tenant SQLite file
- Real-time Engine: integrated WebSocket gateway for instantaneous data broadcasting via pub/sub architecture
- Rule-based Security: granular access control using a custom rule engine that synthesizes SQL WHERE clauses, down to per-field rules
- Embedded Search: first-class support for Full-Text Search (FTS5) and Vector similarity search (sqlite-vec)
- Plugins: supervised Python extensions (JSON-RPC over stdio) with config, lifecycle, logs and invocation
- Replication: master/replica file sync for read scaling
- Standalone Binary: single binary with embedded web frontend, no separate frontend server needed
- Docker Support: official images (API-only and standalone)
Quick Start
The quickest path is the vski command line (Python 3.11+ and Docker):
pip install "vski-sdk[cli]" --extra-index-url https://vski.sh/api/packages/x/pypi/simple/
vski install # pulls vski.sh/x/vski:latest-standalone
vski serve # runs it on http://127.0.0.1:3000
vski doctor # verifies server, auth, docker, image
Open http://127.0.0.1:3000/installer to create the first admin, then:
vski login
vski init my-app # bootstrap a project from the app template
cd my-app && make venv
make iac-plan && make iac-apply
From the template directly
git clone https://vski.sh/x/template.git vski-app
cd vski-app
make dev
Manual setup
# Standalone binary (server + embedded web UI)
wget https://vski.sh/x/platform/releases/download/latest/vski-standalone
chmod +x vski-standalone
./vski-standalone
# Docker: light (API only) or standalone (with web UI)
docker run -p 3000:3000 -v $(pwd)/data:/app/data vski.sh/x/vski:latest
docker run -p 3000:3000 -v $(pwd)/data:/app/data vski.sh/x/vski:latest-standalone
Documentation
- Docs site: https://docs.vski.ai (generated from the wiki, which stays the source of truth)
- Getting Started / Command Line / IaC: the wiki's Start section
- API reference: OpenAPI spec served by the server at
/openapi.json(browsable at/openapi.html), plus the docs in the embedded web UI
Repository layout
| Path | What it is |
|---|---|
vski/ |
The server (Go, SQLite; API, workflows, pipelines, replication) |
tango/ |
The cluster proxy: read/write splitting, write journal, failover with hot promotion |
k8s/ |
The vski-cluster Helm chart (vski StatefulSet + tango + Redis) |
web/ |
The web UI (React), embedded into the standalone binary |
ts-sdk/ |
The TypeScript/JS SDK (@vski/sdk) |
py-sdk/ |
The Python SDK (vski-sdk) and the vski CLI |
examples/plugins/ |
Example Python plugins |
emulator/ |
Playground (experimental) |
docs/ |
Development notes and release notes |
Multi-Tenancy
Reserved database names:
default.db- system metadatastats.db- request statisticsworkflows.db- workflow runs, events and queues (replicated as a passive copy; execution is master-only)
License
See vski/LICENSE for license information.
Links
- Repository: https://vski.sh/x/platform
- Documentation: https://docs.vski.ai
- App template: https://vski.sh/x/template
- Python SDK registry: https://vski.sh/api/packages/x/pypi/simple/
- Issues: https://vski.sh/x/platform/issues
- Releases: https://vski.sh/x/platform/releases