Changelog
Every run should start stronger than the last — so should the product. Here is what shipped, release by release. Reach out at hello@tryultraresearch.com.
- v0.1.0
Ultraresearch v0.1.0
The first release. UltraEvolve moves to an intent-based approach — state the intent and one evolve strategy generalizes across the work — on top of the evolutionary search infrastructure, orchestrator providers, and persistent memory that ship in this build.
- NewUltraEvolve operates on an intent-based approach — one strategy spanning refactoring, development, optimization, mount-only, and read-only instruction tasks.
- NewUltraEvolve deployment mode with a live view of the generate → rollout → verify → score loop, under a per-run compute budget.
- NewFive orchestrator providers: BYOK for OpenAI, Claude Code, and Gemini (Antigravity); ChatGPT Sign-In for Codex; and your existing OpenCode subscription.
- NewPersistent memory that compounds across runs, with three-timescale learning wired into recursive self-improvement.
- NewScale from your laptop to GPU clusters over SSH, with hardware-aware execution across CPU, GPU, and MPS.
- New20+ built-in AI/ML engineering skills in the agentskills.io-compatible format, including autoformalization and literature search.
- ImprovedIntent-based refinements to the evolve strategy for more generalized runs.
- ImprovedSmoother setup and day-to-day usability for the OpenCode provider.
- ImprovedStandard UI polish and bug fixes across the app.