01 — Overview
Internal work produces a lot of long, untidy text: claim files, contracts, support threads, inspection notes. In Chamfer, a Claude step can sit inside any app or Job. Ask the Bench for a claim summary on the review screen or a drafted reply on the refund desk, and it adds the model call, the prompt and the fields the model may read. Usage and billing stay on your own Anthropic account, and you choose the model for each step.
Operators, Chamfer’s agents, can use Claude to work a queue on a named person’s authority. Each Operator is limited to specific tables and actions, every step is logged, and anything outside its limits goes to a person for approval. You can also build Chamfer apps from Claude itself over Chamfer MCP, under the same permissions and Logbook as the Bench.
02 — What you can do
Long-document reading
Send whole claim files, contracts or threads, limited to fields the user may see.
Operators on Claude
Agents work a queue with Claude inside set tables and actions, with one-click revoke.
Build from Claude
Connect Claude to Chamfer MCP and create or edit apps from a conversation, logged.
03 — FAQ
Before you connect Anthropic
Do we need our own Anthropic account?
What can an Operator do on its own?
How is Chamfer MCP different from this integration?
Are prompts and responses stored?
04 — Pairs well with