Orchestrate a fleet of autonomous AI coding agents across multiple repositories
Track and cap how much money or tokens a fleet of agents spends
Watch agent progress and approve decisions through a web dashboard
Recover automatically when an agent's task is interrupted mid-run
| paul-gross/blizzard | 0xallam/my-recipe | 0xhassaan/nn-from-scratch | |
|---|---|---|---|
| Stars | 0 | — | 0 |
| Language | Python | Python | Python |
| Last pushed | — | 2022-11-22 | — |
| Maintenance | — | Dormant | — |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 4/5 | 2/5 | 4/5 |
| Audience | developer | general | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Python 3.12 or newer, installed from a downloaded release wheel rather than a package index.
Blizzard is a platform for managing large groups of AI coding agents working on their own. It handles the parts of the process around the actual coding work: taking in tasks, deciding which agent handles which task, giving each agent its own workspace, checking whether the result is acceptable, delivering the finished work, and recovering cleanly if anything gets interrupted along the way. It also tracks how much each attempt costs in tokens and money, and can set spending limits so a fleet of agents does not run up an unexpected bill. The project is intentionally narrow about what it does. It does not build code, run tests, or review code itself. Instead, it assumes each AI agent is capable enough to figure out how a given codebase builds and tests itself once it is dropped into that codebase's workspace. Because of this, Blizzard has no per project configuration or settings describing how a specific codebase works. Those answers are expected to live in the codebase itself, where the agents can read them directly. Blizzard ships as a single installable package containing everything needed to run it: a hub program that coordinates work, a runner program that supervises individual agents, a command line tool, and a web dashboard for watching progress, all bundled together so no separate JavaScript tooling is required. It uses a simple SQLite database by default, though a more powerful Postgres database can be configured instead. Setting it up involves downloading a prebuilt package from the project's GitHub releases page and installing it with Python's package installer, since there is no separate download index for it yet. After installing, a single command starts the hub, which serves both the coordination service and the web dashboard on a local address. The project also maintains an extensive automated test suite covering realistic end to end scenarios, such as an agent asking a human a question mid task or a human approving a decision before work continues. The README does not mention which license covers the project.
A platform that manages fleets of autonomous AI coding agents, handling task assignment, review, delivery, and spend limits around their work.
Mainly Python. The stack also includes Python, SQLite, Postgres.
The README does not state a license, so its usage rights are unclear.
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.