EN DE
v0.4.4 — Resource Governance Framework (RGF)

Your AI Agents Are
Burning Through Budget.
We Fix That.

RGF is the Resource Governance Framework for Multi-Agent environments — a lightweight proxy layer between your agents and your API budget. OpenAI-compatible, zero custom code, with predictive forecasting, crash recovery, and Governance as Code.

Quick Start → View on GitHub
setup.sh — zsh
$pip install openai
$export OPENAI_BASE_URL="https://api.resgov.silentops.cloud/v1"
→ Budget checked. Stream started. Tokens tracked.
Zero-Risk
Budgets
7+ Frameworks
Open Source
MIT
Open Source
Get Started

Up and Running in 30 Seconds

Self-hosted with Docker. One command, zero config. Works with any OpenAI-compatible client.

quickstart — zsh
$git clone https://github.com/michael-ebering/resgov.git
$cd resgov && docker compose up -d
→ Dashboard: https://api.resgov.silentops.cloud/dash
→ API: https://api.resgov.silentops.cloud/v1
About the Creator

Built for the AI Community

Michael Ebering

Michael Ebering

Built by Michael Ebering — SilentOps

I built RGF because I lived the problem: running multi-agent systems on OpenRouter and watching the API costs spiral out of control. No tool on the market gave me per-agent budget enforcement, crash recovery, and predictive forecasting in a lightweight, self-hosted package — so I built one.

My mission is to make AI infrastructure accessible and sustainable for everyone. I believe that governance shouldn't be an afterthought — it should be as easy as writing a config file. RGF is my way of giving back to the AI community that has given me so much. Open source, self-hosted, and built with care.

🔓 MIT License
🤝 Self-Hosted First
⚡ Transparent LLM Proxy
🌍 Governance as Code
Capabilities

What RGF Does

The complete governance layer for autonomous AI agents — budgets, forecasting, crash recovery, and audit.

🔌

OpenAI-Compatible Proxy

Drop-in replacement for api.openai.com. Change base_url only. Streaming and function calling fully supported. Works with LangChain, CrewAI, OpenAI SDK, Anthropic SDK.

💰

Per-Agent Budgets

Daily, monthly, and total spending limits per agent. Hard deny with clear reasons when limits exceeded — zero upstream costs on denial. Multi-tenant isolation with org-scoping.

⚡

Pre-Commit / Finalize

Reserves pessimistic max cost at stream start (milliseconds lock), refunds difference after completion. No long-held DB locks. No deadlocks under concurrent load.

📈

Predictive Forecasting

Proactively predict when an agent will exhaust its budget based on real spend velocity. Get remaining budget, rate per hour, and estimated time-to-exhaustion — before overruns happen.

🛡️

Crash Recovery

Stuck reservations automatically decay and revert after 5 minutes if an agent crashes mid-stream. Your budget is never orphaned by zombie processes. Self-healing by design.

📊

Real-Time Dashboard

Dark-mode monitoring at /dash — live budget bars, recent bookings, denial counters. Auth-protected. Native Prometheus /metrics endpoint for Grafana.

🔍

Audit Trail + Webhooks

Every request logged with agent, cost, status, timestamp. Paginated API for compliance. HMAC-signed webhook notifications on budget events, denials, and resets.

🔄

Book Any Resource

Non-LLM API calls, web scraping, compute, storage — book them via POST /api/v1/book. Same budget enforcement, same audit trail. Unified control plane.

Declarative Config

Governance as Code

Define budgets, model allowlists, and tool policies in a single .rgf file in your repo. No dashboards, no wizards. Just declarative config, in Git.

Why .rgf? RGF reads a TOML-based config file at startup. All agent budgets, model restrictions, and fail-safe rules live in one file — version-controlled, reviewable, deployable with your stack.

  • Set per-agent daily & monthly limits
  • Restrict allowed models per agent
  • Declare permitted tools and APIs
  • Configure fail-safe action on proxy loss
  • Takes effect on restart — no API call needed
# .rgf — Resource Governance Rules

[global]
currency = "EUR"
fail_safe_action = "deny"

[agents.hermes]
daily_budget = 10.00
max_tokens_per_request = 4096
allowed_models = ["anthropic/claude-sonnet-4-6", "openai/gpt-5.5"]

[agents.research-bot]
daily_budget = 4.00
allowed_models = ["gpt-4o-mini"]
allowed_tools = ["web-scraper", "pexels_search"]

[agents.code-agent]
daily_budget = 6.00
max_tokens_per_request = 8192
allowed_models = ["anthropic/claude-sonnet-4-6"]

[agents.monitor]
daily_budget = 1.50
allowed_models = ["openrouter/deepseek/deepseek-v4-flash"]
Architecture

How It Works

1

Register Agent

POST /api/v1/agents — set daily and monthly budgets per agent. Or define them in your .rgf file.

2

Proxy Request

Agent calls /v1/chat/completions with X-ResGov-Agent-ID header. Zero agent code changes.

3

Budget Enforcement

Reserve max cost → stream → finalize with actual usage. All in milliseconds. No blocked parallel agents.

4

Monitor & Alert

Dashboard + audit trail + webhooks + Prometheus + predictive forecasting. Full visibility into every dollar.

Works With Your Stack

Any framework that supports custom base_url. Zero agent code changes.

API: api.resgov.silentops.cloud · Documentation · ReDoc
Open Source

Free & Open Source

RGF is open source under the MIT License — free for any use with attribution.

Free & Open Source

€0
  • Your infra, your rules
  • Per-agent budget enforcement
  • Predictive cost forecasting
  • Governance as Code (.rgf)
  • Docker deployment, 30 seconds
Get Started →

Get Started

RGF is open source under the MIT License (free for any use with attribution). Self-host with Docker in 30 seconds — or get notified about new releases.

✓ No credit card ✓ Self-hosted ✓ No data leaves your infra
✓

No tracking. No spam. Unsubscribe anytime.

Or: git clone https://github.com/michael-ebering/resgov.git && docker compose up