Hi, we're hiloop

Infrastructure for automated research.

Autoresearch is agents proposing, running, and verifying experiments. hiloop gives them snapshottable, forkable compute where observability is built in. Bring your own harness, or use ours.

Talk to the founders Read the quickstart

Backed byY Combinator

Trusted by

What we build

What autoresearch needs to run.

Provision compute, keep every experiment reproducible, and see what happened across thousands of runs.

Compute your agents provision themselves through the API. A fork branches the run tree and inherits the parent's filesystem from its snapshot: try an idea, abandon it, or restore any snapshot into a fresh sandbox. Secret values are write-only to callers and encrypted at rest, and sandbox bindings fail closed until proof-bound request-time delivery ships.

$ hiloop run -- claude -p "explain this repo"

# → capturing run 01K6Z… · every event fork-tagged

$ hiloop run -- python my_eval.py

# any agent command, yours or ours

✓ llm · tool · stdout, one trace

A homegrown eval script or an off-the-shelf RL loop runs the same way: wrap any command with hiloop run, same provisioning, same traces, no SDK required.

// run 01K72… · fork of 01K6Z… · claude code harness

{"signal":"llm","text":"try a cosine lr schedule, warmup 2k"}

{"signal":"exec","cmd":"python train.py --lr cosine","fork":"01K73…"}

{"signal":"log","stdout":"step 4000 · val_bpb 0.9021"}

{"signal":"llm","text":"worse than parent, restore parent snapshot"}

{"signal":"snapshot","op":"restore","to":"01K6Z…"}

✓ one trace per run · every event fork-tagged

Every run is recorded as one trace, with tokens and estimated cost rolled up per model, queryable with SQL.

Who it's for

Built for teams where performance is the product.

You're building autoresearch

Agents propose, run, and verify experiments against a metric you choose. If you can measure it, an agent can climb it.

You're training models

Every point of accuracy and every dollar of compute counts, and reproducibility across runs is non-negotiable.

Performance is the product

Speed, latency, memory, cost: you iterate constantly, and every experiment needs a record you can trust.