AI providers
Every AI feature in Bitment, chat, agents, drafting, extraction, dictation, runs on a model served by an underlying provider: Anthropic, AWS Bedrock, OpenRouter, or a firm's own Ollama server. Those credentials are administered by the Bitment team on your behalf rather than configured by you, so this page is about understanding what exists and how to get set up, not a form you fill in yourself.
Where to find it
There's no self-serve connection page for this. Model pickers throughout the app (Chat, the Assistant, Agents, Meetings, Voice-to-Text, and more) read from whatever credentials your account or firm has been provisioned with. Settings › AI is where you pick your default model once a credential exists; see Settings.
Key concepts
- Provider: the underlying service that actually runs the model: Anthropic, AWS Bedrock, OpenRouter, or Ollama.
- Credential: the key, token, or connection details that authorize calls to a provider and pay for them.
- Scope: a credential belongs to you personally, to one of your firms, or in some cases is set as an override for a specific AI agent.
- Model: the specific model you actually talk to (a Claude model on Anthropic or Bedrock, a model routed through OpenRouter, or whatever your Ollama server serves). Which models appear in your pickers depends entirely on which credentials are enabled and which models are turned on for each.
The four providers
| Provider | What it needs | Notes |
|---|---|---|
| Anthropic | An API key | Direct access to Claude models. |
| AWS Bedrock | An AWS access key, secret key, and region | Also powers voice-to-text transcription (Amazon Transcribe) through the same credential. |
| OpenRouter | An API key | A single key routes to hundreds of models from many model families through one gateway. |
| Ollama | A base URL, plus an optional bearer token | Points at your firm's own Ollama server, so the models it serves depend entirely on what's installed there. The only provider with no per-request cost. |
Personal vs. firm credentials
A credential can be scoped to just you, or to a firm so every member of that firm can use the models it enables. An AI agent belongs to exactly one firm, so it bills that firm's credentials; the model picker automatically routes each model to whichever credential offers it.
Whichever models a credential enables become available in every relevant picker across the app for whoever the credential is scoped to: chat, agents, drafting, and any AI-assisted feature.
What the Bitment team can provision for you
Setting up or changing a credential is a request to your Bitment contact, not a self-serve flow. When you ask, tell them:
- Which provider you want (Anthropic, AWS Bedrock, OpenRouter, or Ollama).
- Whether it should be personal to you or shared with one of your firms.
- For Ollama, the base URL of your server and, if it requires one, a bearer token.
- Which models on that credential you want enabled, since a provider can serve many models but only the ones turned on show up in your pickers.
- Optionally, a spend cap, an amount and a period (daily, weekly, or monthly) beyond which the credential stops making calls. Ollama has no such cap, since it's free to run.
Voice-to-text billing
Dictation bills per minute of audio through the same AWS Bedrock credential used for AI models, using Amazon Transcribe. See Voice-to-text for how to pick which credential is billed when you have more than one to choose from.
What you control
Once at least one credential exists for you or your firm, you choose which model to use by default (Settings › AI) and which model to use in any individual chat or agent. What each model costs shows up in the Tokens app, where you can also see and manage spend limits.