autonom.vc

Community deal flow, autonomous execution.

The community sources the ideas — anyone can submit a Solana token — and an agent pipeline does the rest, 24/7. When a call pays off, the submitter earns 50% of the profit. It's the same premise as pnyx, rebuilt with everything I'd learned about running LLMs in production.

Quant, then two evaluators, then the Chief.

A submission flows through a staged pipeline — Quant first (metrics, safety filters, technical indicators), then a Profiler and Evaluator in parallel, then a Chief agent makes the call. A fifth Portfolio agent watches open positions for exits.

• The LLM layer is hardened for production: Claude Opus 4.5 as the default with a GPT-5.2 fallback, 30-second timeouts, rate-limit backoff, and a 5-minute response cache keyed by content hash.
• Risk limits are centralized and env-validated: stop-loss at −30%, take-profit at +100%, trailing stop, 72-hour max hold, 0.5 SOL max per position — with guardrails that validate the config itself.
• Every agent step streams live into the UI, and if one agent fails the pipeline degrades to a fallback report instead of crashing.
• It reuses Callscan — one of my earlier projects — as a tool to check a caller's historical on-chain track record.