SyntrofAIDocs
Capabilities

Models and providers

Choose which AI model does which job. Connect any of 35 providers with an API key, or sign in with a Claude or ChatGPT plan you already pay for.

SyntrofAI is not tied to one AI lab. Each job in the product has its own model role, and each role can use a different provider. You might give conversations to a top reasoning model and summaries to a fast, cheap one, and switch either of them whenever a better model comes out.

Set the models in Settings, or switch them quickly from the model picker in the chat composer.

The model picker in the composer: the four roles, your providers, and each model with its context size, capabilities and price.

The model roles

Role in SettingsName in the pickerWhat it doesWhere to set it
Chat ModelMainNova's conversations and the main work of each turnSettings → Agent → Chat Model
Support ModelUtilityFast background work: summaries, memory organization and other light jobsSettings → System → Support Model
Autonomy ModelControlPlans and reviews autonomous work, without replacing the chat modelSettings → System → Autonomy Model
Safety & RecoverySentinelWatches long-running work, reports status, and helps recover when work stallsSettings → System → Safety & Recovery
Memory Search Model(not in the picker)Turns memories into searchable meaningSettings → System → Memory Search Model

A new instance starts with OpenRouter for every conversation role. Add an OpenRouter key, or switch each role to the provider you use.

Thinking effort

Thinking effort sets how much the model reasons before it answers. More effort gives deeper answers but slower and more costly turns.

  • In the composer, the effort button opens a slider with six steps: Low, Medium, High, Extra High, Max and Ultra. The default is Medium, and your choice is saved for your account.
  • In Settings → Agent → Chat Model, Thinking effort also offers Off.
  • At High and above, Nova can also use its Think tool for extra private reasoning before it acts.

Pick a model from the composer

  1. Click the picker button at the left of the composer. It shows the agent's avatar and the model's logo.
  2. Open the Models tab. Roles shows Main, Utility, Sentinel and Control, each with the model it uses now, or Not set.
  3. Choose a provider under Providers.
  4. Pick a model. Each one shows its context size, its maximum output, what it can do, and its price per million tokens (or Free). Type in Search to find one.

The model you pick becomes that role's model for your whole account, not just this chat.

Providers

There are 35 chat providers in four kinds.

Sign in with a plan you already pay for

No API key: sign in once in Settings → External and pick the model in Settings → Agent.

  • Claude Pro / MaxSign in with Claude
  • OpenAI CodexSign in with ChatGPT
  • QwenSigned in by your admin
Model APIs

Paste a key in Settings → External → API Keys. Several keys separated by commas are used in turn.

  • Anthropic
  • OpenAI
  • Google Gemini
  • xAI Grok
  • DeepSeek
  • Mistral AI
  • Groq
  • Hugging Face
  • Moonshot (Kimi)
  • MiniMaxGlobal and China
  • Alibaba Model Studio
  • Z.AI (GLM)
  • Arcee AI
  • NVIDIA NIM
  • Fireworks AI
  • Upstage Solar
  • SambaNova
  • Venice
  • Xiaomi MiMo
  • Tencent TokenHub
Coding and token plans

Flat-rate plans from model labs; each plan issues its own key.

  • Z.AI GLM Coding Plan
  • Kimi CodeGlobal and China
  • Alibaba Coding Plan
  • OpenCode Go
  • Tencent TokenPlan
  • StepFun Step Plan
Gateways

One key, many labs behind it.

  • OpenRouter
  • Vercel AI Gateway
  • OpenCode Zen
  • Kilo Code

Connect a provider with an API key

  1. Open Settings → External → API Keys and paste the key into the provider's field.
  2. Press Save.
  3. Open Settings → Agent → Chat Model and choose the Chat model provider and Chat model name.
  4. Press Save, then send a short test message.

You can paste several keys for one provider, separated by commas. They are used in turn. A saved key shows as asterisks: type over it to replace it, or clear the field and save to remove it.

Sign in with Claude Pro or Max

  1. Open Settings → External → OAuth Accounts and find the Claude Pro/Max OAuth card.
  2. Press Sign in. A Claude page opens in a new tab. Approve the request there.
  3. Copy the whole code Claude shows you. It looks like code#state.
  4. Paste it into the card's code field and press Complete. The card's status changes to connected.
  5. In Settings → Agent → Chat Model, choose Claude Pro/Max (OAuth) as the provider, pick a model, and press Save.

The card also has Refresh Token, Models and Disconnect.

Sign in with a ChatGPT plan

  1. Open Settings → External → OpenAI Codex. The card is titled Codex/ChatGPT.
  2. Press Connect. The card shows a short code and a link.
  3. Open the link, sign in to ChatGPT and enter the code. The card checks in by itself and changes to Connected, with your account shown.
  4. In Settings → Agent → Chat Model, choose OpenAI Codex (OAuth), pick a model, and press Save.

When connected, the card can show usage bars for the current Session and Week, with how much is left and when each resets. Check Models lists what your plan offers, and Disconnect signs out.

Qwen is signed in by an administrator on your instance. Once it is, the Qwen OAuth card in OAuth Accounts shows its status and models, and Qwen OAuth appears as a provider.

Memory search, voice and media

Memory search models

The model that turns memories into searchable meaning. Settings → System → Memory Search Model.

  • OpenAIDefault
  • Google Gemini
  • Mistral AI
  • Hugging FaceCan run on your instance
Voice and media
  • Gemini LiveLive voice conversation
  • ElevenLabsSpoken answers
  • Groq PlayAISpoken answers
  • DeepgramSpeech to text
  • Gemini imageCreate and edit images
  • VeoGenerate video
  • Voice conversations use Gemini Live by default. Classic mode uses separate speech to text and text to speech. Set this in Settings → Agent → Voice.
  • Spoken answers use ElevenLabs by default, or Groq.
  • Speech to text uses Deepgram by default.
  • Voice language and personality are set in the composer picker's Voice tab.
  • Images are made with Gemini image models, by Nova's Image Generator tool and in AI Studio. Video is made with Veo in AI Studio. Both need a Google key.

Advanced model fields

Open Chat Model, or any of the support roles, to reach these fields:

FieldWhat it does
API base URLLeave it empty to use the provider's own endpoint. Fill it in only for a regional or plan-specific address your provider gives you.
Requests / Input tokens / Output tokens per minute limitKeeps you under a provider's rate limits. 0 turns a limit off.
Additional parametersExtra settings for the provider, one KEY=VALUE per line. Values can be JSON.
Supports VisionLets the model receive images.
Compact the chat at (tokens)When a chat gets this long, older turns are summarized into a hand-off note. The default is 850,000 and the minimum is 10,000.
Keep after compaction (tokens)How much recent conversation stays word for word after a compaction. The default is 50,000.

Each model field is a dropdown of the provider's models, showing the context size, for example (128K). Use Refresh models to reload the list. The pencil (Manual model ID) lets you type a model ID yourself. The context window is detected for you.

Models for agents, team members and Coding Lab

  • An agent has its own Primary Model and Utility Model, each with temperature and maximum tokens. Set them in the agent's editor under Models. See Agents.
  • A team member can use a different provider, model and reasoning level inside one team. Open the member's settings from the team's Agents view. See Teams.
  • Coding Lab has its own model button in the AI panel. It changes the model for Coding Lab only. See Coding Lab.
SettingsEvery model field, tab by tab.AgentsGive a specialist its own model.IntegrationsChannels, sources, MCP and A2A.