Skip to content

gojogourav/Vortex

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
cmd
cmd
 
 
internal
internal
 
 
k8s
k8s
 
 
vortexMobile
vortexMobile
 
 
.gitignore
.gitignore
 
 
Dockerfile
Dockerfile
 
 
README.md
README.md
 
 
go.mod
go.mod
 
 
go.sum
go.sum
 
 

Repository files navigation

VoxDeploy

AI-powered CI/CD pipeline that automatically fixes broken builds and lets you control your entire infrastructure in any language.

When your GitHub Actions workflow fails, VoxDeploy catches the webhook, reads the error logs with Gemini AI, generates a precise git diff, and opens a pull request — all without human intervention. Your engineers can also control Kubernetes clusters, Docker containers, and query metrics by speaking or typing in Hindi, Spanish, Japanese, or any other language via the companion mobile app.

Screenshots

Mobile App

Overview Voice Control PRs

Settings

Architecture

CI Auto-Fix Pipeline

flowchart TD
    A([GitHub Actions\nworkflow fails]) --> B[Webhook received\nHMAC signature verified]
    B --> C[Download + extract\nworkflow logs ZIP]
    C --> D[Filter error lines\nFAIL · error · undefined]
    D --> E[Gemini 2.5 Flash\nLogParser\nreturns file paths as JSON]
    E --> F{Files found?}
    F -- No --> G([Pipeline aborted\nno targets identified])
    F -- Yes --> H[Fetch broken files\nfrom GitHub API\nin parallel goroutines]
    H --> I[Lingo.dev\nTranslate fix instruction\nHindi → English]
    I --> J[Gemini 2.5 Flash\nFixGenerator\ntemperature 0.0\nreturns unified diff]
    J --> K[Fuzzy diff applier\nbestAnchor search ±15 lines\nUnicode normalization]
    K --> L{Diff applied\nsuccessfully?}
    L -- No --> M([DiffApplyError\nmetric incremented])
    L -- Yes --> N[GitHub API\nCreate fix branch\nvoxdeploy/fix-abc1234]
    N --> O[Push patched files\nto fix branch]
    O --> P([PR opened\nReady for review])

    style A fill:#ff4466,color:#fff
    style P fill:#0f7b6c,color:#fff
    style G fill:#9b9a97,color:#fff
    style M fill:#9b9a97,color:#fff
Loading

Mobile Voice Pipeline

flowchart LR
    A([Hold mic\nbutton]) --> B[expo-av\nRecords audio\nM4A / AAC]
    B --> C[Release button]
    C --> D[Groq Whisper\nLarge v3\nSpeech → text\nauto-detects language]
    D --> E[Lingo.dev\nTranslates to English\nHI · ES · JA · ZH · DE]
    E --> F{Mode}
    F -- K8s --> G[k8s.InterpretCommand\nkeyword matching]
    F -- Docker --> H[docker.InterpretCommand\nkeyword matching]
    F -- Grafana --> I[grafana.InterpretCommand\nPromQL queries]
    G --> J[client-go\nKubernetes API]
    H --> K[Moby SDK\nDocker API]
    I --> L[Prometheus\n/api/v1/query]
    J --> M([Result displayed\nin app])
    K --> M
    L --> M

    style A fill:#37352f,color:#fff
    style M fill:#0f7b6c,color:#fff
Loading

System Overview

graph TB
    subgraph Mobile["📱 Mobile App (React Native)"]
        APP[VoxDeploy App]
        VOICE[Voice Screen]
        CTRL[Control Screen]
        DASH[Overview Dashboard]
    end

    subgraph Backend["🖥️ VoxDeploy Backend (Go)"]
        GW[Gateway]
        WH[WebhookHandler]
        K8SH[K8sCommandHandler]
        DCH[DockerCommandHandler]
        GRH[GrafanaCommandHandler]
        VTH[VoiceTranscribeHandler]
    end

    subgraph AI["🤖 AI Services"]
        GEMINI[Gemini 2.5 Flash\nFix generation]
        GROQ[Groq Whisper v3\nSpeech-to-text]
        LINGO[Lingo.dev\nTranslation]
    end

    subgraph Infra["⚙️ Infrastructure"]
        K8S[Kubernetes\nclient-go]
        DOCKER[Docker\nMoby SDK]
        PROM[Prometheus\nMetrics]
        GITHUB[GitHub API\nWebhooks + PRs]
    end

    APP --> GW
    VOICE --> VTH
    CTRL --> K8SH
    CTRL --> DCH
    CTRL --> GRH
    DASH --> GRH

    GW --> WH
    WH --> GEMINI
    WH --> LINGO
    WH --> GITHUB

    VTH --> GROQ
    K8SH --> LINGO
    K8SH --> K8S
    DCH --> LINGO
    DCH --> DOCKER
    GRH --> LINGO
    GRH --> PROM
Loading

Fuzzy Diff Applier

flowchart TD
    A[Raw diff from Gemini] --> B[ParseDiff\nextract FileDiff structs]
    B --> C[For each hunk]
    C --> D[bestAnchor\nfind longest non-blank\ncontext line in hunk]
    D --> E[Search ±15 lines\naround stated OldStart]
    E --> F{Anchor found?}
    F -- No --> G([Error: hunk anchor\nnot found])
    F -- Yes --> H[Compute realStart\n= anchorLine - anchorOffset]
    H --> I[Apply hunk lines\nContext · Added · Removed\nwith Unicode normalization]
    I --> J{All hunks\napplied?}
    J -- No --> C
    J -- Yes --> K([Patched file\nready to push])

    style G fill:#ff4466,color:#fff
    style K fill:#0f7b6c,color:#fff
Loading

What it does

Automatic CI repair — A GitHub Actions failure fires a webhook. VoxDeploy downloads the logs, filters the error lines, uses Gemini to identify which files are broken, fetches those files from the repo, generates a unified diff fix at temperature 0, applies the diff with fuzzy line matching, and opens a PR on a voxdeploy/fix-* branch. The whole pipeline runs in under 60 seconds.

Multilingual infrastructure control — The mobile app (React Native / Expo) lets you send natural language commands to Kubernetes, Docker, and Grafana in any language. Commands pass through Groq Whisper for speech-to-text and Lingo.dev for translation before being interpreted and executed against your cluster.

Multi-repo support — Register multiple repositories with independent GitHub tokens, kubeconfigs, Docker hosts, and Grafana instances. Each webhook is routed to the correct per-repo configuration.

Tech stack

Component Technology
Backend Go 1.22
AI fix generation Google Gemini 2.5 Flash
Speech-to-text Groq Whisper Large v3
Translation Lingo.dev
Mobile app React Native (Expo)
Kubernetes client k8s.io/client-go
Docker client Moby SDK
Metrics Prometheus + Grafana
CI integration GitHub Actions webhooks

Project structure

Vortex/
├── cmd/
│   ├── server/
│   │   └── main.go                 # Entry point
│   └── internal/
│       └── api/
│           ├── handlers.go         # Webhook, K8s, Docker, Grafana handlers
│           ├── voice.go            # Groq Whisper transcription
│           ├── k8shandler.go       # Scale / restart / list REST endpoints
│           ├── register.go         # Multi-repo registration
│           └── signature.go        # GitHub HMAC verification
├── internal/
│   ├── diff/
│   │   ├── apply.go                # Fuzzy hunk applier
│   │   ├── parse.go                # Unified diff parser
│   │   ├── types.go                # FileDiff / Hunk / Line types
│   │   └── diff_test.go            # 30+ tests
│   └── metrics/
│       └── metrics.go              # Prometheus counters + histograms
├── cmd/internal/
│   ├── llm/
│   │   ├── client.go               # Gemini client (LogParser + FixGenerator)
│   │   └── prompts.go              # System prompts
│   ├── github/
│   │   ├── logs.go                 # Log download + ZIP extraction
│   │   ├── files.go                # File content fetcher
│   │   └── git.go                  # Branch create / file update / PR open
│   ├── k8s/
│   │   ├── client.go               # Kubernetes client factory
│   │   └── commands.go             # Command interpreter + executor
│   ├── docker/
│   │   ├── client.go               # Docker client factory
│   │   ├── containers.go           # Container operations
│   │   └── command.go              # Command interpreter
│   ├── grafana/
│   │   ├── client.go               # Grafana/Prometheus client
│   │   ├── commands.go             # NL query interpreter
│   │   ├── stats.go                # In-memory metrics reader
│   │   └── query.go                # PromQL executor
│   ├── lingo/
│   │   └── client.go               # Lingo.dev translation client
│   └── store/
│       └── store.go                # Per-repo config store
├── k8s/                            # Kubernetes manifests
│   ├── deployment.yaml
│   ├── service.yaml
│   ├── rbac.yaml
│   └── namespace.yaml
├── docs/
│   └── screenshots/                # Add your screenshots here
└── vortexMobile/                   # React Native app
    ├── app/
    │   ├── _layout.tsx
    │   ├── index.tsx               # Overview / metrics dashboard
    │   ├── control.tsx             # K8s + Docker + Grafana control
    │   ├── voice.tsx               # Voice + text command input
    │   ├── prs.tsx                 # PR history + pipeline steps
    │   └── settings.tsx            # Repo registration
    ├── components/
    │   ├── theme.ts
    │   ├── NotionCard.tsx
    │   ├── PropertyRow.tsx
    │   ├── CalloutBlock.tsx
    │   └── ResultBlock.tsx
    ├── hooks/
    │   └── Usevoicecommand.ts
    └── services/
        └── api.ts

Setup

Prerequisites

  • Go 1.22+
  • A GitHub repository with Actions enabled
  • kubectl configured with access to your cluster
  • Docker daemon running

Environment variables

GITHUB_TOKEN=github_pat_...        # Personal access token (repo scope)
WEBHOOK_SECRET=your-secret         # Must match GitHub webhook configuration
GEMINI_API_KEY=AIza...             # Google AI Studio — free tier available
LINGO_API_KEY=lingo_sk_...         # lingo.dev — free tier available
GROQ_API_KEY=gsk_...               # console.groq.com — free, no credit card
PORT=8080                          # Optional, defaults to 8080

Run locally

git clone https://github.com/yujiblack/Vortex.git
cd Vortex

PORT=8081 \
GITHUB_TOKEN="..." \
WEBHOOK_SECRET="secret" \
GEMINI_API_KEY="..." \
LINGO_API_KEY="..." \
GROQ_API_KEY="..." \
go run cmd/server/main.go

Expose to GitHub (for local development)

ngrok http 8081
# Copy the https URL and set it as your GitHub webhook URL

Configure GitHub webhook

  1. Go to your repository → Settings → Webhooks → Add webhook
  2. Payload URL: https://your-ngrok-url.ngrok.io/webhook
  3. Content type: application/json
  4. Secret: same value as WEBHOOK_SECRET
  5. Events: select Workflow runs

Register a repository

curl -X POST http://localhost:8081/repos/register \
  -H "Content-Type: application/json" \
  -d '{
    "repo": "owner/repo-name",
    "github_token": "github_pat_...",
    "webhook_secret": "your-secret",
    "kubeconfig_b64": "'$(kubectl config view --context your-context --minify --flatten | base64 -w 0)'",
    "docker_host": "unix:///var/run/docker.sock",
    "grafana_url": "http://localhost:9090",
    "grafana_token": ""
  }'

Deploy to Kubernetes

# Create namespace and RBAC
kubectl apply -f k8s/namespace.yaml
kubectl apply -f k8s/rbac.yaml

# Create secrets
kubectl create secret generic voxdeploy-secrets \
  --from-literal=GITHUB_TOKEN="..." \
  --from-literal=WEBHOOK_SECRET="..." \
  --from-literal=GEMINI_API_KEY="..." \
  --from-literal=LINGO_API_KEY="..." \
  --from-literal=GROQ_API_KEY="..." \
  -n voxdeploy

# Deploy
kubectl apply -f k8s/deployment.yaml
kubectl apply -f k8s/service.yaml

Mobile app

cd vortexMobile
npm install
npx expo install expo-av expo-file-system react-native-safe-area-context

# Update BASE_URL in services/api.ts to your server URL

# Run in Expo Go (typed commands, no mic)
npx expo start

# Run with full mic support
npx expo run:ios
# or
npx expo run:android

App screens

Overview — Live metrics: webhooks caught, PRs opened, success rate, AI latency, diff errors. Auto-refreshes every 10 seconds.

Control — Natural language commands to Kubernetes, Docker, or Grafana. K8s pod listing, deployment scaling, pod restarts, crashing pod detection, Docker container management, Prometheus metric queries.

Voice — Hold the mic button and speak in any language. Groq Whisper transcribes, Lingo.dev translates to English, the command executes. Supports EN, HI, ES, JA, ZH, DE. Typed input fallback with example phrases filtered by language.

PRs — AI-generated summary of pipeline health, stats, link to open PRs on GitHub, visual walkthrough of the 6-step fix pipeline.

Settings — Register new repositories with independent credentials per repo.

API endpoints

Method Path Description
POST /webhook GitHub Actions webhook receiver
POST /k8s/command Natural language K8s command
POST /docker/command Natural language Docker command
POST /grafana/command Natural language metrics query
GET /grafana/stats Raw pipeline metrics JSON
GET /k8s/deployments List deployments for a repo
POST /k8s/scale Scale a deployment
POST /k8s/restart Restart a deployment
POST /voice/transcribe Audio → transcript via Groq Whisper
POST /repos/register Register a new repository
GET /repos List registered repositories
GET /health Health check
GET /metrics Prometheus metrics

How the diff applier works

The biggest engineering challenge is that Gemini occasionally miscounts line numbers when generating diffs. The fuzzy hunk applier solves this with bestAnchor — it finds the longest non-blank context line in each hunk and searches ±15 lines around the stated position to find the real location. Blank lines are terrible anchors (they appear everywhere), so the algorithm picks the most unique line in the hunk instead. Unicode normalization handles curly vs straight apostrophes so comment lines never cause false mismatches.

AI states:  @@ -30,13 +30,13 @@
Actual pos: func at line 37

bestAnchor: "func (p *Product) ApplyDiscount..." at line 36
anchorOffset = 7 (7th line into the hunk)
realStart = 36 - 7 = 29 ✓

WARNING: hunk line number corrected: stated 30, actual 29 (off by -1)

Metrics

Metric Description
voxdeploy_webhooks_received_total CI failures caught
voxdeploy_fixes_generated_total Diffs successfully generated
voxdeploy_prs_opened_total Pull requests opened
voxdeploy_fix_failed_total Pipeline failures at any stage
voxdeploy_diff_apply_errors_total Diff parse or apply failures
voxdeploy_ai_generation_seconds Gemini latency histogram
voxdeploy_pipeline_duration_seconds Full webhook-to-PR latency
voxdeploy_files_patched_per_fix Files changed per fix

Example voice commands

Language Command Action
🇬🇧 English "Show all pods" List all pods
🇬🇧 English "Scale voxdeploy-test to 3" Scale deployment
🇬🇧 English "Restart voxdeploy-test" Rolling restart
🇮🇳 Hindi "सभी pods दिखाओ" List all pods
🇮🇳 Hindi "voxdeploy-test को 3 replicas पर scale करो" Scale deployment
🇮🇳 Hindi "crashing pods दिखाओ" Show failing pods
🇮🇳 Hindi "AI की speed कितनी है" Query AI latency
🇪🇸 Spanish "Mostrar todos los pods" List all pods
🇪🇸 Spanish "Escalar voxdeploy-test a 3" Scale deployment
🇪🇸 Spanish "Dame un resumen" Pipeline summary
🇯🇵 Japanese "すべてのpodを表示" List all pods
🇯🇵 Japanese "voxdeploy-testを再起動" Restart deployment
🇯🇵 Japanese "成功率はどのくらいですか" Query success rate

License

MIT

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages