Notes on other agent platforms

01 · The watchlist

Systems worth learning from

Each entry is what the system is, and what Valor took from studying it.

Claude Managed AgentsAnthropic · hosted harness

A pre-built agent loop in managed cloud containers, driven over an API. Intended to take over execution later; that offload waits on hosted memory and a hook-equivalent API.

Paperclipopen source · control plane

Org charts, per-agent budgets, and approval gates over any agents. Source of the budget hard-stop, goal-aligned tracing, and immutable audit-log ideas being adopted natively.

HermesNous Research · headless framework

A headless multi-tenant agent across twenty chat platforms, with agent-curated memory and self-improving skills. The strongest argument for indexed session stores, and the skills ecosystem worth watching.

OpenHumantinyhumans · desktop app

A one-user second brain that normalizes everything into readable Markdown trees. The strongest argument for a legible file store: half of why Valor's memory runs two stores.

Perplexity ComputerPerplexity · digital worker

A vision-language agent that drives a real browser the way a person would. The benchmark for visual web work, and the reason an orchestrator needs an API it can call.

Pi & Gondolinminimal core · micro-VM sandbox

A widely-adopted minimal agent core and a standalone micro-VM sandbox with the cleanest isolation seam evaluated: deliberately not adopted, because Valor's dedicated machine is its isolation boundary.

OmnigentDatabricks · control plane

A credible open-source agent control plane that looked like a toy in second-hand accounts and wasn't: the standing reminder to verify maturity claims at the source.

jcodecoding-agent runtime

A fast-moving single-author coding runtime: high capability, no integration seam. The pattern for when to watch rather than adopt.

02 · The comparisons

Head-to-head studies

Full write-ups of the in-house strategic analyses, one page per matchup.