Execution layer for AI agents. The AI writes a JSON recipe, the Go engine runs it on your machine. File ops, HTTP requests, scripts, chaining — deterministic, every time.
What it is
Every AI agent has the same problem — they can reason, plan, generate code, but they can't reliably execute anything on your actual machine. Vibra-Ingenn is a bolt-on execution layer. It doesn't replace your AI stack, it gives it hands.
Connect it to Claude, Cursor, VS Code, Windsurf, or any MCP-compatible tool. The AI writes a JSON recipe using an 8-field task schema. The Go runtime executes it deterministically. Same input, same output, every time.
Setup — 4 lines in your MCP config:
// Add to your AI tool's MCP settings { "vibra-ingenn": { "command": "vibra.exe", "args": ["mcp"] } }
That's the whole setup. No Python, no Docker, no npm install, no configuration files. Download the binary, add those 4 lines, go.
The schema
Every task in every recipe uses the same structure. The AI fills these fields, the engine executes them.
| Field | What it does |
|---|---|
| id | Unique task name within the recipe |
| trigger | What fires it — manual, event:prev_success, schedules |
| action_type | What kind of work — see below |
| target | Where it's directed — file path, URL, script, model |
| payload | Action-specific data, supports {{template_vars}} |
| auth | Reference name into the encrypted secrets store |
| output | Where results go — store, pass to next, write to file, log |
| next | Chain to next task, or branch on success/failure |
Action types
Real example
This recipe hits the GitHub API, writes a report, and confirms. Three tasks, chained.
{
"workflow_id": "github-trending-go",
"tasks": [
{
"id": "fetch",
"trigger": "manual",
"action_type": "http_request",
"target": "https://api.github.com/search/repositories?q=language:go&sort=stars",
"payload": { "method": "GET" },
"output": { "pass_to_next": true },
"next": "write"
},
{
"id": "write",
"trigger": "event:prev_success",
"action_type": "file_op",
"target": "C:\\reports\\trending_go.txt",
"payload": { "operation": "write", "content": "{{prev_output}}" },
"next": "done"
},
{
"id": "done",
"trigger": "event:prev_success",
"action_type": "notify",
"target": "stdout",
"payload": { "message": "✓ Report saved." }
}
]
}
Those are real numbers from an actual run. The file write took half a millisecond.
Pricing