iterate-ai 0.2.0

Autonomous ML iteration agent: a Supervisor briefs experiments, a coding agent writes and runs real notebook cells against your data, and the winner ships as a runnable notebook. Multi-LLM backend (local Ollama to cloud), persistent memory, bounded autonomy.

iterate

Autonomous research-aware iteration agent for ML models and LLM prompts.

pip install iterate-ai

# your CSV, your target column, your metric. LLM runs on local Ollama ($0)
# or any OpenAI-compatible endpoint. Full setup: Quick start below.
iterate run --data examples/churn_tabular/data.clean.csv --target Churn --metric f1

iterate runs an autonomous experiment loop on your ML problem. In v0.2 the agent writes and runs its own training code, cell by cell, in a live Jupyter kernel: a Supervisor reads the run history and briefs one experiment, a coding agent executes it against real cell outputs and real tracebacks, a Summarizer distills every finished notebook so the next one inherits what worked and what failed. Every submission is scored on a sealed holdout, every attempt persists in memory, and the winner ships as a runnable notebook. 391 unit tests across 31 files run in CI on every push.

v0.2 today	On the roadmap
The agent writes + runs its own training code (any model, any preprocessing) in a local kernel or an isolated e2b sandbox	Interactive CLI: pause, mid-run chat, resume, token streaming (v0.3)
Multi-agent core: Supervisor briefs, coding agent executes cell by cell, Summarizer carries knowledge across notebooks	Researcher + Critic specialists; agent picks the metric + starting model (v0.4)
A deterministic guard stack converts weak-model waste: grounded briefs, duplicate hashing, dead-end transfer, floor submissions	LLM prompt iteration (v0.5), vision transfer learning (v0.6)
Winner ships as a runnable notebook (hypothesis, staged cells, real outputs, dead ends labeled, findings); persistent memory across runs	Cost-to-serve recommendations (v0.7), MCP auto-discovery of data + context (v0.9)

Why I built this

I kept seeing the same failure mode on small AI teams. A model or a prompt ships, and under delivery pressure nobody iterates on it again, so it sits in production for months while baselines move on. Experiments get re-run because nobody wrote down why they failed the first time. Teams pay frontier-model prices because nobody checked whether a cheaper model with a better prompt would do the job. iterate is the institutional memory, research desk, and experiment runner those teams don't have time to build.

How this gets built: WORKFLOW.md (the method) · DECISIONS.md (every call I made against the AI's default) · BUILD_LOG.md (the daily trail)

---

The full pitch

Every YC batch ships 200+ AI startups with 2-3 engineer teams. Under shipping pressure, two things break: nobody re-iterates models against new baselines, and LLM prompts sit in production for months untouched. Engineers re-run failed experiments because nobody logged why. Teams pay GPT-5 prices because nobody tested whether Haiku + better prompting would do the job at 1/50th the cost.

AutoML brute-forces. Experiment trackers only log. Prompt evals only evaluate. AIDE iterates Kaggle problems once. iterate is being built as the system that runs an autonomous, literature-aware, memory-persistent improvement loop on ML models, DL/vision models, AND LLM prompts in production, optimizing for the best model you can actually afford to serve. That is the v1.0 vision; the releases below get there one dial at a time.

---

Status

v0.2 released: the agent writes and runs its own training code. In v0.1 the agent picked estimators from an allow-list; in v0.2 it works like an engineer in a notebook, and the winner ships as one.

Agent-first: the autonomous loop landed at v0.1 (Week 3), not as a late-stage add-on. Two dials turn release to release: the inputs you must give shrink (toward one-sentence input) and the problem types grow (tabular, then prompts, then DL/vision).

Week	Phase	Status
0	Scaffolding + scope lock	done
1	Foundation: schemas + LLM client (tool-calling) + config + CLI	done
2	Tabular execution substrate: BenchmarkTarget + data adapter + ModelTarget + model factory + local executor	done
3	The agentic loop: Proposer + Orchestrator + Terminator + Memory + CLI, first autonomous tabular run (v0.1)	done
4-5	Sandboxed code-gen + the multi-agent cell-by-cell system (Supervisor, coding agent, Summarizer, pulled forward from the original v0.4 plan) + notebook deliverable + live progress + graceful Ctrl-C (v0.2)	done
6	Full interactive CLI: pause, mid-run chat, resume + token streaming (v0.3)	next
7	Researcher + Critic specialists; agent picks the metric + starting model (v0.4)	planned
8	PromptTarget: agentic prompt iteration (v0.5)	planned
9	DLModelTarget: vision transfer learning, validated on a local RTX 4050 (v0.6)	planned
10	Cost-constrained recommendation + serving profile + iterate cost (v0.7)	planned
11	Infer features/target from the data + a description (v0.8)	planned
12	MCP discovery: find the data/code itself (v0.9)	planned
13	Multi-backend benchmark + Streamlit chat UI + demos (v0.10)	planned
14	Full minimum-viable-input + polish + launch (v1.0)	planned

---

What v0.2 does

You give it a prepared CSV, the target column, and a metric. The agent does the rest: builds its own baseline, then runs one briefed experiment per iteration, cell by cell, against a sealed holdout it never sees.

What a live run looks like:

coder[iter-04]: cell 6 ok (1.7s, 4/300s budget)
coder[iter-04]: cell 7 error: NameError: name 'Xb_cat' is not defined
coder[iter-04]: cell 8 ok (0.1s, 5/300s budget)
agent loop: iteration 4 'Model Swap - XGBoost' -> f1=0.6312

                        Run summary
 iter   model                              f1   delta vs baseline
 base   baseline                       0.5676                   -
    1   Baseline Model                 0.6251             +0.0575
    2   Class Weight Balancing         0.6251             +0.0575
    3   Hyperparameter Tuning          0.6279             +0.0603
    4   Model Swap - XGBoost  <- best  0.6312             +0.0636
    ...
best: Model Swap - XGBoost (f1=0.6312, +0.0636 vs baseline)

Each iteration is a real R&D session, not a script dump:

1. The Supervisor compresses everything tried so far into a two-line brief: the banked best (exact config, threshold, components), the known dead ends, and exactly ONE new move to try.
2. The coding agent rebuilds the carried best, applies the brief's one change, measures it like-for-like on a validation split, and submits only what it can defend. Errors are debugged from real tracebacks, cell by cell.
3. The Summarizer digests the session (what helped, what hurt, the takeaway), so run 7 knows what run 3 learned.
4. The harness scores the submission on the sealed holdout and saves the notebook immediately. Ctrl-C keeps everything already earned.

The harness is the moat, not the model. Every fact in a brief is machine-derived from the actual banked code, never LLM recall. A stack of deterministic guards catches the failure modes weak models actually produce: briefs that re-commission already-banked work, submissions byte-identical to earlier ones, briefed changes that never reached a line of code, sessions that die without submitting (a floor submission banks automatically). Each guard exists because a live forensic run demonstrated the failure it prevents; the stack was validated across 21 instrumented runs on a local 12B model, which ties its all-time best score inside the guarded loop.

Deliverables. The winner is exported as best.ipynb: a runnable, annotated notebook of the actual winning session (hypothesis, staged cells with their real outputs, dead ends labeled, findings). --notebooks all keeps one notebook per iteration: the full research journey. Every experiment also persists in .iterate/memory.db, so the next run builds on this one.

---

Quick start

Local-first. $0. No API keys required.

# 1. Install Ollama + a local model (one-time)
brew install ollama
ollama pull gemma4:12b         # the model v0.2 was validated on
ollama serve                   # background server at localhost:11434

# 2. Install iterate (pulls scikit-learn / XGBoost / LightGBM)
pip install iterate-ai         # "iterate" was taken on PyPI; the command is still `iterate`

# 3. Prepare a tabular CSV (your standard ML data cleaning) and run
iterate run --data train.clean.csv --target churn --metric f1

The first run offers a one-time setup wizard (backend, model, compute, install consent); after that, flags override saved defaults per run.

# Run the generated code in an isolated cloud sandbox instead of locally:
iterate run --data train.clean.csv --target churn --metric f1 --compute e2b

# Use a cloud LLM backend (aliases: groq, together, deepseek, openai):
iterate run --data train.clean.csv --target churn --metric f1 \
            --backend groq --model llama-3.3-70b-versatile --api-key "$GROQ_API_KEY"

# Seed the baseline from an existing notebook/script (read as text, never executed):
iterate run --data train.clean.csv --target churn --metric f1 \
            --source baseline_notebook.ipynb --baseline 0.78

# Bound the whole run; keep every iteration's notebook:
iterate run --data train.clean.csv --target churn --metric f1 \
            --until 30m --notebooks all

Useful flags: --max-iterations, --patience, --until (wall-clock bound), --notebooks best|all|none, --compute local|e2b, --install/--no-install (package-install consent), --think (reasoning mode for the coder, Ollama only), --fresh (archive memory, start a new chapter), --spec (the v0.1 allow-list path, kept as the fast lane). Full reference: iterate run --help

Where things land: .iterate/runs/<run_id>/best.ipynb (the runnable winner), notebooks/ (with --notebooks all), best.json (config + score sidecar). Code-path winners ship as notebooks by design; --spec winners also save best_model.joblib.

Safety boundaries: your --source file is read as text, never executed. The generated code runs locally only with your consent (the setup wizard asks), or fully isolated with --compute e2b. The holdout labels never enter the kernel; scoring happens host-side.

Note on the one-line form. The iterate "improve our churn baseline" experience, where the agent discovers the data, baseline, and metric itself, is the v1.0 vision, not v0.2. Today you pass --data/--target/--metric explicitly; the inputs shrink release by release (see the roadmap). Auto-discovery, prompt + vision targets, and cost-constrained serving are on the roadmap, not shipped yet.

---

Three target families (the v1.0 shape)

Target	What it iterates on	Status
ModelTarget	Trains a tabular model, scores it on a sealed holdout	shipped (v0.1, code-gen in v0.2)
DLModelTarget	Transfer-learns a vision model, scores it	planned (v0.6)
PromptTarget	Runs an LLM prompt against an eval set, scores outputs	planned (v0.5)

All inherit from BenchmarkTarget. Same iteration loop, different execution path. (LLMs are prompt-iteration only; we don't fine-tune foundation models.)

---

Pluggable data + tools via MCP (Week 12, v0.9)

iterate will use Model Context Protocol (MCP) servers as its discovery layer: filesystem, Postgres, Notion and friends, so adding a data source is config, not code. The discovery workflow (agent introspects your tables, past experiments, and notebooks, then pauses for your gap-fill) lands at v0.9. Today the data interface is a prepared CSV, deliberately: the loop had to be proven before the input surface grows.

---

Architecture

src/iterate/
├── core/                 # the reasoning engine
│   ├── agent_loop        # v0.2 loop: Supervisor briefs -> coder runs -> Summarizer digests
│   ├── supervisor        # strategist: grounded briefs + deterministic no-op guards
│   ├── coder             # cell-by-cell coding agent on a live stateful kernel
│   ├── summarizer        # per-experiment digest (cross-notebook knowledge transfer)
│   ├── codegen           # code-gen contract, session preamble, floor submission
│   ├── orchestrator      # v0.1 spec-path loop (--spec)
│   ├── proposer          # spec-path proposer + dataset profiling
│   ├── reconstructor     # rebuild a baseline from --source (text only, never executed)
│   ├── memory            # persistent experiment store (sqlite)
│   ├── scoring           # sealed-holdout scoring, shared by both paths
│   └── terminator        # deadline / patience / max-iterations gates
├── targets/              # BenchmarkTarget protocol + tabular ModelTarget
├── adapters/
│   ├── data/             # csv loading + profiling
│   ├── models/           # estimator registry (spec path)
│   └── compute/          # LocalKernel + E2BKernel (Jupyter), runners, sandbox
├── deliver/              # runnable .ipynb rendering (sessions, leaderboards)
├── llm/                  # pluggable backends: native Ollama client + one
│                         #   OpenAI-compatible client (Groq/Together/Deepseek/OpenAI/vLLM)
├── prompts/              # every prompt in one yaml, versioned with the code
└── schemas/              # Pydantic types

The LLM is plug-and-play; the harness does the lifting. The same loop runs on a local 12B or a cloud 70B. The bet (see the infra-over-model A/B in EVAL_LOG.md): a good enough harness makes weak local models perform like much bigger ones, and the guard stack is what closed that gap.

---

Comparison with existing tools (the v1.0 target)

Capability	AutoML (DataRobot/H2O)	W&B / MLflow	Braintrust / LangSmith	AIDE	iterate
Iterates ML models autonomously	✓	✗	✗	✓	✓ shipped
Agent writes its own training code	✗	✗	✗	✓	✓ shipped
Persistent memory across sessions	✗	log only	✗	✗	✓ shipped
Bounded autonomy (deadline / patience)	✗	✗	✗	partial	✓ shipped
Auditable reasoning trail (runnable notebooks)	✗	✗	✗	basic	✓ shipped
Iterates LLM prompts	✗	✗	eval only	✗	planned v0.5
Iterates DL / vision models	partial	✗	✗	partial	planned v0.6
Literature-aware proposals	✗	✗	✗	partial	planned v0.4
Cost-to-serve-aware optimization	✗	✗	✗	✗	planned v0.7
Auto-discovers data + context (MCP)	✗	✗	✗	partial	planned v0.9
Open-source	mostly ✗	MLflow yes	✗	✓	✓

Known limits are documented honestly in LIMITATIONS.md; the evaluation trail lives in EVAL_LOG.md.

---

License

MIT. The framework is open-source. Adapters for proprietary data sources can be built on top.

---

Author

Anthony Rodrigues: GitHub