Mira is an agent-oriented LLM with instinctive reasoning compression and safe recursive self-improvement loops. Designed for long-horizon tasks, tool use, and self-play fine-tuning with eval gating. Results in.evalresults/: Terminal-Bench 2.1 Not AGI. Agent-oriented LLM with heuristic compression. Recursive self-improvement is offline and eval-gated. Tool use requires guardrails. Apache-2.0