{"slug": "hillclimb-open-source-alphaevolve-style-search-with-claude-code-and-codex", "title": "Hillclimb: Open-source AlphaEvolve-style search with Claude Code and Codex", "summary": "Hillclimb launched as an open-source CLI and modular Python framework for building, testing and benchmarking hillclimbing algorithms driven by coding agents such as Claude Code and Codex. The framework separates a problem, defined by a verifier script (verifier.sh) and optional starting solution (solution.py), from a climber, a search algorithm composed of five modules — selector_policy, operator_policy, operators, tuner and memory — with catalog climbers greedy, openevolve and gepa. Installed via `uv tool install hillclimb` or `pip install hillclimb` on Python 3.12+, it runs searches in the background under a user-set compute budget and keeps the best result in runs/<run-id>/searches/<search-id>/best/.", "body_md": "**Hillclimbing on verifiable rewards.**\n\nUse Claude Code and Codex to autonomously search for python programs that optimize a score you define.\n\n[Website](https://hillclimb.sh?utm_medium=readme) ·\n[Docs](https://docs.hillclimb.sh?utm_medium=readme) ·\n[Quickstart](https://docs.hillclimb.sh/quickstart?utm_medium=readme) ·\n[Python SDK](https://docs.hillclimb.sh/python?utm_medium=readme) ·\n[CLI reference](https://docs.hillclimb.sh/cli?utm_medium=readme) ·\n[Blog](https://docs.hillclimb.sh/blogposts?utm_medium=readme) ·\n[Discord](https://hillclimb.sh/discord?utm_medium=readme)\n\n`hillclimb` is a CLI and modular Python framework for building, testing and benchmarking hillclimbing algorithms powered by coding agents. It provides reusable components and a common way to evaluate different search strategies across different problems — from speeding up code and solving optimisation problems to improving prediction accuracy. The goal of `hillclimb` to create a framework to enable discovery of better autoresearch methods, rather than build a single best autoresearcher.\n\nThere are two main parts: the [problem](https://docs.hillclimb.sh/problems?utm_medium=readme) and the [climber](https://docs.hillclimb.sh/climbers?utm_medium=readme). A problem defines what counts as better through a verifier script, `verifier.sh`, and optionally includes a starting solution, `solution.py`. A climber is a search algorithm that drafts initial solutions and then decides which solutions to build on and what to try next. Start with built-in problems and climbers, or write your own ones.\n\n[`hillclimb run`](https://docs.hillclimb.sh/cli/commands/run?utm_medium=readme) starts the search in the background and returns control of your terminal. Within the compute budget you set, the climber dispatches headless coding agents to write, debug and improve solutions. Your verifier scores each attempt, and `hillclimb` keeps the best result.\n\nA `hillclimb` search running visualised through `hillclimb watch`:\n\n```\nuv tool install hillclimb                  # or: pip install hillclimb (Python 3.12+)\nhillclimb connect claude                   # or codex, pi, openrouter; add --agent dummy to any run for no LLM at all\n\nhillclimb init                             # hillclimb.yaml, problems/, runs/, climbers/ in this folder\nhillclimb problem get heilbronn-convex-13  # the verifier is the problem; description.md is the brief\nhillclimb climber get greedy               # the climber, as Python you can read and edit (climbers/greedy/policy.py)\nhillclimb verify heilbronn-convex-13       # score the floor solution\n\nhillclimb run heilbronn-convex-13 --budget 10m\nhillclimb watch                            # live coding agents and scores (also: chart, tree)\nhillclimb stop --all\nhillclimb summit                           # copy the best solution.py into this folder\n```\n\nEvery search also keeps its best in `runs/<run-id>/searches/<search-id>/best/`.\nThe [walkthrough](https://docs.hillclimb.sh/walkthrough?utm_medium=readme) goes through each step.\n\nA climber is five modules, run in this order at every step of a search. The\nengine ships none of the first two: the catalog's climbers (`greedy`,\n`openevolve`, `gepa`) are Python files `hillclimb climber get` copies into your\nfolder, to read and edit; the other three are built in or a file of your own.\n\n| Module | Decides | Built in / in the catalog | \n|---|---|---|\n| `selector_policy` | which candidate the next attempt builds on | `Best` ,`MapElites` (catalog) | \n| `operator_policy` | which operator to apply to it | `Greedy` (catalog) | \n| `operators` | how one attempt is made: the prompt the coding agent gets | `draft` ,`debug` ,`improve` ,`ensemble` | \n| `tuner` | which parameter values to try on a candidate | `random` ,`optuna` | \n| `memory` | what carries over from one search to the next | `files` ,`none` | \n\nTogether they are one block of config, in a run spec or as the folder's run\ndefault in `runs/config.yaml` (`hillclimb climber get` pins one there):\n\n```\nclimber:\n  selector_policy: climbers/openevolve/policy.py:MapElites   # which candidate to build on next\n  operator_policy: climbers/greedy/policy.py:Greedy           # which operator to use on it\n  operators: [draft, debug, improve, crossover.py:Crossover]\n  tuner: optuna\n  memory: files\n  params: {num_drafts: 5}\n```\n\nEvery slot takes a `.py` file (`file.py:Class`), `package.module:Class`, or a\nbuilt-in's name. A fetched climber is one file that builds the whole block\n(`climber: climbers/greedy/policy.py`). Or compose in Python:\n\n``` python\nfrom hillclimb import Budget, Climber, Problem, catalog\n\ngreedy, openevolve = catalog.module(\"greedy\"), catalog.module(\"openevolve\")\nproblem = Problem(\"heilbronn-convex-13\")\nbudget = Budget(wall_clock=\"10m\", evaluations=40)\nclimber = Climber(selector_policy=openevolve.MapElites(num_drafts=5), operator_policy=greedy.Greedy())\n\nclimber.search(problem, budget=budget)\nclimber.best, climber.history, climber.to_frame()\n```\n\nA catalog climber that imports a library beyond hillclimb (`openevolve`, `gepa`) names it in a\n`requirements.txt` beside its `policy.py` — `pip install -r climbers/gepa/requirements.txt`, as\n`climber get` says; `tuner: optuna` is the engine's and needs `'hillclimb[optuna]'`.\n\n`climber.start(...)` opens the same search to drive by hand, one `climber.step()` at a\ntime, and `Problem(name, score=my_function, ...)` defines a problem from a scoring function.\nRunnable scripts are in [`examples/`](https://github.com/rebase-energy/hillclimb/tree/main/examples).\n\nCompare two head to head:\n\n```\nhillclimb climber get openevolve --no-default\nhillclimb run heilbronn-convex-13 --climber climbers/greedy/policy.py --climber climbers/openevolve/policy.py   # one search each; --parallel-searches 3 runs three of each\nhillclimb experiment report <run-id>\n```\n\n| From first run to following and resuming a search. | Define your own problem and write its verifier. | \n| The modules a climber is built from. | Compare climbers on the same problem. | \n| Every example problem and its best known value. | Every `hillclimb` command and its flags. | \n| Build and run searches from Python. | Blog posts on hillclimbing. | \n\nQuestions, results, ideas: [join the Discord](https://hillclimb.sh/discord?utm_medium=readme).", "url": "https://wpnews.pro/news/hillclimb-open-source-alphaevolve-style-search-with-claude-code-and-codex", "canonical_source": "https://github.com/rebase-energy/hillclimb", "published_at": "2026-10-16 17:37:31+00:00", "updated_at": "2026-10-16 17:50:17.744494+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "ai-research", "artificial-intelligence"], "entities": ["Hillclimb", "Claude Code", "Codex", "Anthropic", "OpenAI", "greedy", "openevolve", "gepa"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/hillclimb-open-source-alphaevolve-style-search-with-claude-code-and-codex", "markdown": "https://wpnews.pro/news/hillclimb-open-source-alphaevolve-style-search-with-claude-code-and-codex.md", "text": "https://wpnews.pro/news/hillclimb-open-source-alphaevolve-style-search-with-claude-code-and-codex.txt", "jsonld": "https://wpnews.pro/news/hillclimb-open-source-alphaevolve-style-search-with-claude-code-and-codex.jsonld"}}