The autonomous AI system that builds, wires, heals, and evolves itself -- with zero API keys required.
At the core of SolarPunk is the AUTO_GENESIS loop -- a perpetual cycle that discovers what the system needs, builds it, wires it into the existing network, and then evaluates the results. Every cycle, the system grows smarter, more connected, and more resilient.
Unlike traditional software that requires humans to write every line of code, AUTO_GENESIS uses a local AI model (Ollama) to generate new engines on demand. No cloud APIs. No API keys. No external dependencies. The system is sovereign.
The system scans the mycelium/ directory for all Python files. Each file is an "engine" -- a self-contained module that reads from data files, processes information, and writes results. The scanner identifies which engines exist, what they consume, and what they produce.
Once all engines are mapped, the system identifies gaps: data files that are produced but never consumed (orphan outputs), data files that are expected but never produced (hungry inputs), and functional capabilities that are missing. Currently tracking 8 orphan outputs and 60 hungry inputs that AUTO_GENESIS works to fill each cycle.
Using a local Ollama model, the system generates new Python engines to fill discovered gaps. Each generated engine follows a strict template: it must have a run() function, declare its input and output data files, and handle errors gracefully. The generated code is written to disk and immediately available for the next cycle.
The live wire protocol automatically connects engines by matching outputs to inputs. If Engine A writes data/report.json and Engine B reads data/report.json, a wire is created between them. This produces the 5,476 active wires that form the system's nervous system. Wires are classified as zero-secret (no API keys needed) or keyed (requires external credentials).
The nanobot heal system continuously monitors engine health. If an engine crashes, produces malformed output, or fails its health check, the nanobots attempt automatic repair: re-generating the engine, resetting its state file, or quarantining it if repair fails. The immune system tracks threat patterns and builds memory to prevent recurring failures.
The chimera evolution engine merges successful patterns from different engines to create new hybrid capabilities. If a revenue engine and a content engine both perform well, their patterns can be fused into a "chimera" that handles revenue-generating content. Mutations are tracked on a leaderboard, and the best mutations survive across generations.
SolarPunk was designed with a radical constraint: it must be able to run without any API keys, tokens, or external credentials. This is the Zero-Secret Army -- currently 283 engines that operate entirely on local computation, file I/O, and Ollama-based AI inference.
This design choice matters because:
The engine interconnection topology is inspired by mycelium -- the underground fungal networks that connect trees in a forest. Each engine is a node. Each data file is a nutrient pathway. The system exhibits biological properties:
data/pheromone_map.json to coordinate behaviordata/immune_memory.jsonThe 366 engines are organized into functional categories, each serving a distinct role in the organism:
All of this is coordinated by the OMNIBUS -- the master orchestrator that runs the full AUTO_GENESIS cycle. Currently at version 38, the OMNIBUS:
Every claim on this page can be independently verified:
ls mycelium/*.py | wc -l on the public repositorydata/live_wire_report.json in the repodata/chimera_evolution_report.json