Pricing
Why we don't resell tokens
AgentLoom runs on your own Anthropic key. Anthropic bills you directly; we charge a flat fee for orchestration. That split is the product working as intended.
June 11, 2026 · The AgentLoom team
There are two costs in agentic delivery, and most products blur them into one bill. Model usage is what Anthropic charges for tokens. Orchestration is everything that turns those tokens into merged, verified work: composing the right team, running the lifecycle, surviving CI waits and review rounds, repairing main, feeding follow-ups back into the backlog. AgentLoom keeps the two separate on purpose: you bring your own Anthropic key, Anthropic bills you directly for the model, and AgentLoom charges a flat monthly fee for the orchestration.
What bring-your-own-key looks like in practice
You connect an Anthropic API key — or a Claude Code OAuth token if your team already has a Claude subscription — during onboarding.
The credential is encrypted at rest and injected into each run's isolated environment; AgentLoom never proxies or marks up your model traffic.
Revoking is yours: remove the credential and runs stop dispatching. Enterprise can route model traffic through self-hosted Bedrock instead.
Data posture follows the key: model calls run under your direct agreement with Anthropic, which matters to every security reviewer who asks where the code goes.
Incentives are the product
A vendor that resells tokens earns margin on consumption — every retry, every bloated context, every wasted session is revenue. We didn't want to build a company whose best quarter is your least efficient one. With a flat orchestration fee, the incentive flips: AgentLoom wins when runs are efficient, when checkpoints mean nothing is recomputed, when a CI wait costs zero because the run is asleep, and when review feedback lands in one revision instead of four. Harness quality becomes our margin — which is exactly the thing you're paying us to be good at.
What you see, and where it stops
Your dashboard shows the model spend your own key reports, per run, totaled against your plan's monthly cap on orchestrated agent work. No two issues cost the same — review churn, merge conflicts, and retries land in the total honestly. When you hit the cap, runs pause; upgrade or wait for the next period. And the idle states cost nothing by construction: adding seats, reading the dashboard, waiting for CI, waiting for your own reviewers — none of it consumes anything.
Seats are free on every plan for the same reason. Headcount was never the cost driver; execution is. Pricing should say what the product believes — ours believes the value is the orchestration, so that's the only thing we charge for.