optim-agent

Agentic system optimization with coding agents. Automate the iterative parameter-tuning work of an algorithm engineer.

English | 简体中文 | 日本語 | 한국어 | Français | Deutsch | Español | Português | Русский

optim-agent lets Claude Code / Codex / OpenCode tune real system parameters by reading your code, proposing trials, and recording measured objective results. Use it when your system exposes configurable parameters and a measurable objective. It combines what each parameter means with what the trial history shows, then proposes the next configuration to evaluate. Objective evaluations remain authoritative: optim-agent proposes values, validates them against the declared space, records outcomes, and falls back to safe sampling when an agent reply is invalid.

Models	Systems	Research
Training, architecture, and RL experiments	Inference, latency, cost, control, and decision rules	Quant signals, simulations, and scientific workflows

Why optim-agent

• Semantic proposals - coding agents reason over parameter meanings, study context, and observed outcomes instead of treating every dimension as an anonymous coordinate.
• Small-budget leverage - useful when evaluations are expensive and classical surrogates are still data-starved.
• Agent CLI upside - proposal quality can improve as the underlying coding agents improve, such as moving from GPT-5.5 to GPT-5.6, without changing your optimization code.
• Auditable decisions - JSON/SQLite studies retain configurations, outcomes, states, context, and optional agent rationale.
• Bounded execution - the agent only proposes values; optim-agent validates them against the declared space, and invalid output falls back to safe sampling.

Install

Install the Codex skill:

$skill-installer install https://github.com/Optim-Agent/optim-agent

Install the Claude Code plugin:

claude plugin marketplace add Optim-Agent/optim-agent && claude plugin install optim-agent@optim-agent

Install the Python package:

# Stable release from PyPI
python -m pip install optim-agent

# Latest source from GitHub
python -m pip install "optim-agent @ git+https://github.com/Optim-Agent/optim-agent.git"

Requires one authenticated agent CLI on PATH: claude, codex, or opencode.

Quickstart

import optim_agent as oa

def objective(trial):
    threshold = trial.suggest_float(
        "threshold", 0.05, 0.95,
        context="decision threshold; higher values trade recall for precision",
    )
    budget = trial.suggest_int(
        "budget", 10, 200, log=True,
        context="compute or operating budget; larger values may improve quality",
    )
    return evaluate_system(threshold=threshold, budget=budget)  # domain code

study = oa.create_study(
    direction="maximize",
    sampler=oa.AgentSampler(
        backend="claude",  # or "codex" / "opencode"
        effort="high",
        context="maximize system quality under a strict operating-cost budget",
        history=5,
        explicit_reasoning=True,
        qualitative_notes=True,
    ),
    storage="study.json",  # optional: persist and resume
    summarize=True,  # optional: agent-written result summary after the last trial
)
study.optimize(objective, n_trials=20)
print(study.best_value, study.best_params)
print(study.summary)  # the summary agent's narration of the finished study

Optional context gives domain meaning to the study and parameters. Provide it study-wide on AgentSampler(context=...), per parameter on suggest_*(..., context=...), or both.

Where It Applies

Area	Parameters optim-agent can tune	Example objective
Model training	learning rates, architectures, augmentation, regularization	validation quality, compute, robustness
Inference and serving	quantization, batching, decoding, caching, routing	quality, latency, throughput, cost
Quantitative research	signal windows, thresholds, rebalance rules, risk controls	walk-forward return, drawdown, turnover
Reinforcement learning and decisions	objective weights, exploration schedules, environment settings, policy thresholds	return, safety, sample efficiency
Scientific workflows	simulation inputs, solver settings, experimental controls	fit, error, runtime, resource use
Black-box systems	any bounded categorical, integer, or continuous configuration	scalar objective score

For reinforcement learning, optim-agent tunes the system around the learning loop; it does not replace the policy-learning algorithm.

Optimization Trajectory

This seed-0 Branin trace compares TPE and GPT-5.5 under the same 10-trial budget, with incumbent objective values after each trial. It is a trajectory illustration; aggregate benchmark results and reproduction commands follow.

Optimizing Math Functions without Context: Branin-2D and Ackley-5D

Hard-function agents receive no supplied task context: only generic x1...x5 parameter names, numeric bounds, and trial history. Runs use 10 trials over five seeds; Random and TPE are unchanged baselines.

Top-tier Agents

method	mean best Branin ↓	mean best Ackley-5D ↓
Random	5.008	19.639
TPE	11.395	18.843
GPT-5.5	1.326	3.960
Opus-4.8	0.398	0.061
Sonnet-5	3.850	0.143
Kimi-K3	2.082	0.907
Minimax-M3	0.970	0.574
GLM-5.2	3.609	15.023

The pinned models are gpt-5.5, claude-opus-4-8, claude-sonnet-5, kimi-k3, MiniMax-M3, and glm-5.2. Opus-4.8 reaches the Branin optimum on average and has the strongest five-seed Ackley mean.

OpenCode Agents (Free)

method	mean best Branin ↓	mean best Ackley-5D ↓
Random	5.008	19.639
TPE	11.395	18.843
Big-pickle	4.734	15.951
DeepSeek-V4-Flash	4.410	4.608
Nemotron-3-Ultra	16.051	18.459
MiMo-v2.5	3.682	15.597

OpenCode-hosted models require no paid model API. The free pool rotates; this refresh pins opencode/big-pickle, opencode/deepseek-v4-flash-free, opencode/nemotron-3-ultra-free, and opencode/mimo-v2.5-free. DeepSeek V4 Flash has the strongest free-model Ackley mean, while MiMo-v2.5 has the strongest free-model Branin mean.

Tuning ResNet-based Image Classifier: MNIST and CIFAR-10

The classification benchmark compares Random, Optuna TPE, GPT-5.5 w/ context, and GPT-5.5 w/o context over five seeds (0..4) and 10 trials. The context condition receives natural-language study and parameter descriptions; the no-context condition receives only bounds and trial history.

For classification, the primary metric emphasizes fast improvement:

cumulative_best_so_far_error = sum(best_test_error_so_far_at_i for i in 1..10)

Lower is better.

method	MNIST cumulative error ↓	MNIST final error ↓	CIFAR-10 cumulative error ↓	CIFAR-10 final error ↓
Random	9.174	0.648%	278.920	25.072%
TPE	7.166	0.580%	279.936	25.596%
GPT-5.5 w/ context	5.668	0.506%	220.994	21.322%
GPT-5.5 w/o context	8.910	0.632%	281.466	25.960%

GPT-5.5 w/ context reduces cumulative best-so-far error by 20.9% relative to TPE on MNIST and by 20.8% relative to Random on CIFAR-10. Without context, it is 24.3% worse than TPE on MNIST and 0.9% worse than Random on CIFAR-10. The gap includes both semantic parameter information and earlier access to agent-guided proposals.

Both examples/mnist.py and examples/cifar10.py tune learning rate, batch size, weight decay, label smoothing, three stage widths, three stage depths, and four dropout controls. MNIST adds translation and rotation; CIFAR-10 uses crop padding and flip probability.

Tuning Q-learning Controllers: Acrobot-v1 and LunarLander-v3

This CPU-only Gymnasium benchmark tunes a discretized Q-learning controller for Acrobot-v1 and LunarLander-v3. Each method runs 20 trials over five seeds (0..4); the objective is mean evaluation return, so higher is better. The runner parallelizes across seeds and within each HPO study via --workers. The GPT-5.5 arms use high modeling effort and the last 5 trials of history. The