HyperResearch turns Claude Code into persistent research agent
HyperResearch is an open-source Claude Code skill harness that wraps research into a multi-stage workflow with a persistent, searchable SQLite-backed knowledge store, breadth-first source collection, depth investigations, and adversarial fact-checking. It installs with a single command, adds Claude Code skills and subagents, and can use crawl4ai plus authenticated browser sessions to reach sources that standard web search may miss. The project also claims strong performance on DeepResearch Bench, but that benchmark result should be treated as author-reported unless independently replicated.
This looks more like a research operating system than a single agent, and that’s the right direction if the goal is durable knowledge accumulation instead of one-off answers.
- –The persistent vault is the real differentiator: session-to-session reuse can compound value if the indexing and retrieval are solid.
- –The 16-step / layercake-style pipeline is aimed at process quality, not just model quality, which is where most deep-research agents fail.
- –Authenticated crawling broadens coverage, but it also increases risk around source quality, policy boundaries, and reproducibility.
- –The benchmark claim is attention-grabbing, but without third-party replication it’s marketing, not settled fact.
- –The Claude Code-only constraint narrows the audience, though it may also keep the system simpler and more reliable.
DISCOVERED
3h ago
2026-04-29
PUBLISHED
4h ago
2026-04-29
RELEVANCE
AUTHOR
heisdancingdancing