mmvfirm/universal-earned-autonomy-layer — explained in plain English
Analysis updated 2026-05-18
Add graduated permission controls to an AI agent that writes to a database or calls external APIs.
Enforce hard limits on irreversible actions like deletions or payments regardless of agent trust level.
Track a signed, tamper evident audit trail of every agent decision for compliance review.
Run the same control plane across multiple app replicas that share one Postgres database.
| mmvfirm/universal-earned-autonomy-layer | 1lystore/awaek | 47cid/wp2shell-lab | |
|---|---|---|---|
| Stars | 13 | 13 | 13 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 4/5 | 2/5 | 4/5 |
| Audience | developer | vibe coder | researcher |
Figures from each repo's GitHub metadata at analysis time.
Full production use needs a real key vault and identity provider, local demo runs with make dev and make demo.
Universal Earned Autonomy Layer is a Python library that manages how much freedom an AI agent is allowed to have when taking real actions. Instead of trusting an agent from the start, it makes the agent climb an eight step ladder for each combination of task type and authority level, earning more trust only after it produces clean decisions that a human has checked. High risk or irreversible actions are capped by fixed ceilings, so no amount of accumulated trust ever lets an agent skip human review for the most sensitive tasks. This is a rewrite of an earlier version. The first version only reasoned about what authority an agent claimed for itself and returned a status message, but it never actually stopped the agent from proceeding. This new version binds every decision to a signed token, checks that token against a policy enforcement point before anything real happens, and verifies each agent's identity with cryptographic signatures rather than trusting a self reported name. The project includes a signed and tamper evident audit log, support for running multiple copies of the system safely against a shared Postgres database, and tools for measuring how trustworthy an agent has proven itself using statistics rather than guesswork. It ships with a command line tool, an HTTP API with login support, a software development kit for building agents, Docker setup, and automated tests, all runnable with a handful of make commands like make dev, make test, and make demo. The README is unusually direct about what is and is not included. The core safety logic, enforcement, and audit trail are described as tested and ready to run. Genuinely production grade deployment work, such as storing private keys in a proper hardware vault, connecting to a real login provider, and running a security review, is explicitly left for whoever deploys the system. This project would suit developers and teams building AI agents that take real world actions, such as writing to a database or calling external services, and who want a way to gradually and safely expand what those agents are allowed to do. It is released under the Apache 2.0 license.
A Python control plane that lets AI agents earn more autonomy over time, enforcing hard limits on risky actions instead of just advising against them.
Mainly Python. The stack also includes Python, Postgres, SQLAlchemy.
Apache 2.0: free to use, modify, and distribute, including commercially, as long as you keep copyright and license notices.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.