Operators read carefully and ask hard questions. Here are the answers we already give people on the fit call — written down, before you have to ask.
Missing a question? Send it. We’ll answer it personally and add it here if it’s likely to come up again.
Six clusters. Twelve answers. One honest read.
The vault, the routine traces, the meeting transcripts, the staged drafts — all stay on your device. The only thing that talks to a network is the model, and you choose which model, which provider, and which key.
When you approve a send, that one outbound message goes through your own integration (Gmail, Slack, etc.) — also keyed to your account, not ours.
Crash reports if you opt in (off by default). The license check, which is a single signed token call. That’s it. No telemetry, no usage analytics, no “anonymized” memory. We don’t have a database that mirrors your work.
Yes — bring your own key is the default. Anamora speaks to Anthropic, OpenAI, or any OpenAI-compatible endpoint. The recommended model for the assistant turn is Claude Opus 4.7 or GPT-5.4; routines and triage default to a smaller, cheaper model.
Your keys live in the OS keychain on your install. They never reach our servers.
Three ways, your call. Bot mode — Anamora joins your Zoom, Meet, or Teams call as a named participant and records on its own. System audio — Anamora records the audio of whatever’s playing on your machine, no bot needed. Voice memo — for the meetings you couldn’t record, dictate the recall and Anamora tags it with lower confidence.
Yes. Every extracted item carries a timestamp, a speaker, and the exact line in the transcript. Click any commitment in the post-meeting view and you walk back to the moment that produced it.
You move the vault folder to a Mac mini in your office or a small Linux box you own. Same binary, same memory, same approvals — the runtime doesn’t care where the files live. Most operators stay on the laptop forever; the always-on tier is for the few who want routines firing while the lid is closed.
Your vault is plain Markdown in a folder you own. Your routines are YAML. Your runs are append-only logs. Everything keeps working without us — the runtime can be locked to a model provider you keep paying directly, and the file format is documented and open.
If the worst happens, the export is the directory you already have.
Yes — through the always-on deployment. Run Anamora on a Mac mini or a small server, share the vault between operators, and reach the board from anywhere on Tailscale. Approvals are per-operator; the memory is shared.
We don’t recommend more than three people sharing a vault — the cost of one assistant trying to model a too-large team starts to outweigh the benefit.
Anamora is a one-person operation today. Each install is hand-shaped to your stack and language during week three, and three a month is the rate at which I can actually do that well solo. I’d rather grow the cohort slowly than ship generic installs.
No. The board, chat, and approvals are designed for an operator, not an engineer. You may end up editing a YAML file by week six because your business changed and the operator pack needs a tweak — but that’s a fifteen-minute thing, and we walk you through the first one.
Most installs land between $4,000 and $9,000 for the four-to-six week build, depending on your stack complexity and how bespoke the operator pack needs to be. After handoff there’s a small monthly runtime fee — $120/mo for the desktop install, $240/mo for the always-on tier. Model usage is billed directly by your model provider.
The exact number for your install is disclosed in the scoping doc before you commit to anything.
Yes. If by the end of the supervised pilot week (stage 7) the system isn’t earning its keep, we refund the install fee in full. The vault stays on your machine — yours to keep. We’ve never had to do this, but we want you to know the door isn’t locked.
If your concern isn't on this page, the contact form goes straight to a person on the team.