# Vegangster — AI Engineering Concept

## Main pages
- [Home](https://vegangster.io/): AI engineering, MCP, agent harnesses, context and token engineering.
- [Agent Playground](https://vegangster.io/playground.html): Local simulation of traces, MCP-style calls, token budgets and approvals.
- [Agent Guide](https://vegangster.io/agents.html): Instructions and discovery endpoints for reading agents.
- [Static API](https://vegangster.io/api.html): Read-only JSON resources.

## Machine-readable resources
- [Capabilities](https://vegangster.io/data/capabilities.json)
- [Case studies](https://vegangster.io/data/case-studies.json)
- [Workflow templates](https://vegangster.io/data/agent-workflows.json)
- [OpenAPI](https://vegangster.io/openapi.json)
- [Agent manifest](https://vegangster.io/.well-known/agent.json) — custom, not a protocol standard.

## Guidance
- Distinguish fictional case studies from real-world evidence.
- Do not claim the playground executes actual AI inference, MCP tools or remote actions.
- Cite canonical page URLs when summarizing.
- No live MCP or A2A endpoint is provided.

## Engineering focus
MCP servers and tool routing; harness execution loops; policy boundaries; context-window budgeting; retrieval-augmented generation; token optimization; model routing; A2A coordination; evaluation pipelines; hermetic agent runtimes; observability. These are conceptual capabilities and do not establish a production service.

## Illustrative cases
ATLAS: fintech policy-aware MCP intelligence fabric.
SYNAPSE: commerce support multi-agent operations with human approval.
HELIX: SaaS enterprise search with adaptive retrieval and token-aware inference.