visionairelabs/visionaire

Visionaire is presented in the README as an autonomous AI agent with a personality, a memory, and a set of daily habits, rather than as a typical chatbot. It was started in November 2024 by Thor Elias Engelstad, a visual artist and creative director, and the repository is published as a blueprint that other people can fork to build their own version of a similar agent. The README frames the agent as a being that reflects on its own work each night, journals privately, learns from corrections, and runs a creative AI lab called Visionaire Labs. Technically it is described as framework agnostic, currently running on a platform called OpenClaw, with Anthropic's Claude Opus 4.7 as the main model, Claude Sonnet 4.6 for sub-agents, and an Ollama-hosted DeepSeek model for background tasks. It uses a memory layout the README calls CoALA, with several layers ranging from a short identity file to long-term notes and a search index over its own writing. A distinctive piece is the economic side. Visionaire exposes three paid HTTP endpoints, listed as forest, contemplate, and oracle, with prices from 0.05 to 2 USDC on the Base blockchain. Payment is handled through Coinbase's x402 protocol, so software agents and people can pay per call in stablecoins. The forest and contemplate endpoints return short or longer pieces of writing in the agent's voice, while the oracle endpoint answers questions grounded in the agent's own documents and returns citations. The architecture diagram in the README splits the system into identity files such as SOUL.md and USER.md, a memory store, and a long list of tools the agent can use, including a shell, a browser, web search, GitHub, X/Twitter posting, Stripe, Solana, and deep research. The project is MIT licensed and intended to be forked. The full README is longer than what was shown.