ResGov
ResGov (Resource Governance Framework)
A proxy broker that caps LLM agents' daily token and cost budgets and blocks a request before it reaches the provider
Medium risk
We rate an entry medium when the tool runs code, makes network calls or reads project files. Check what exactly it does before installing.
Why this level
- Acts as a proxy for real LLM provider API keys
- The admin token and admin endpoints give full control over limits and keys
Install
Manual install
git clone https://github.com/michael-ebering/resgov.git
cd resgov
cp .env.example .env
docker compose up -dRuns the broker locally, dashboard at localhost:8080/dash.
This is third-party code. Review the repository files before installing.
What it does
ResGov runs as a separate service via docker compose and is dropped in as the base_url for an OpenAI, Anthropic or OpenRouter-compatible client, for example in CrewAI or LangChain. Limits are set in a declarative .rgf TOML file: a daily budget, allowed models and allowed tools per agent. During streaming the service reserves the request's maximum possible cost before the response starts and refunds the difference once it ends, and it blocks a call before it reaches the provider with a 403 response if the budget is exceeded. A separate /api/v1/book endpoint tracks and caps spend on non-LLM resources like paid search or scraping, and the service also offers budget-exhaustion forecasting and Prometheus metrics.
Who it is for. For teams running several autonomous LLM agents who want to hard-cap their API spend before the bill arrives after the fact.
Good fit when
- Several agents share one provider API key and you need a separate daily budget per agent
- There is a risk an agent loops and burns through the budget overnight
- You need to cap not only LLM calls but also paid tools like search or scraping
Not a fit when
- The LLM provider's built-in billing alerts are enough for you
- You cannot run a separate Docker service as a proxy in front of the provider
Example request
Set a 4 euro daily budget for the research-bot agent and allow it only the gpt-4o-mini modelLimitations
The project is young, pre-1.0, and the roadmap still includes horizontal scaling via Redis and Helm charts. The default storage is SQLite in WAL mode, built for a single instance rather than distributed load. It needs separate deployment and administration: your own admin token, HTTPS and backups have to be set up yourself following DEPLOYMENT.md.
How to disable. Point the client's base_url back to the provider directly and stop the service with docker compose down.
Security check
- Acts as a proxy for real LLM provider API keys
- The admin token and admin endpoints give full control over limits and keys
README in short
The README explains the idea of a cost-governance layer for agents on top of MCP and A2A, lists features: a transparent proxy to providers, atomic cost reservation during streaming, budget configuration via a .rgf file, non-LLM resource tracking, multi-tenancy and budget forecasting. It gives a docker compose quick start, CrewAI and LangChain connection examples, the response format on a budget block, an API reference and an architecture diagram. It separately notes this is an independent single-author project unaffiliated with any employer. MIT license.
FAQ
Do I need to rewrite my agent's code?
No, the README shows connecting it by changing a single base_url in OpenAI-compatible SDK clients like CrewAI and LangChain.
What happens when a budget is exceeded?
The request is blocked before it reaches the provider, returning a 403 response describing the limit and current spend.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent