LLMs

How LLMs work

Every agent has an LLM model (llm_id) that controls what the agent says and does. The agent and the LLM model are separate resources so you can share one brain across multiple agents — update the system prompt once and every agent using that model picks up the change immediately.

Agent (voice, first message, call behaviour)
  └── LLM model (system prompt, provider, model, tools, knowledge bases)

When you create an agent, OneInbox automatically creates an LLM model with sensible defaults and returns its id as llm_id.

Providers

OneInbox supports several LLM providers. Platform providers work out of the box — no credentials needed.

Provider	Models	Credential required
OpenAI	`gpt-4o`, `gpt-4o-mini`, `gpt-5`	No — platform-provided
Shisa	`shisa-ai/shisa-v2.1-llama3.3-70b`	No — platform-provided
Anthropic	`claude-sonnet-4-5`, `claude-haiku-4-5`	Yes — add an integration
Groq	`llama-3.3-70b-versatile`, `gemma2-9b-it`	Yes — add an integration
OpenRouter	Curated subset — see model list in playground	No — platform-provided
Custom (WebSocket)	Your own model server	Yes — WebSocket URL

Set a system prompt

The system prompt defines the agent’s persona, goals, and rules. Update it at any time without rebuilding the agent:

curl -X PATCH https://api-tokyo.oneinbox.ai/v1/models/<llm_id> \
  -H "Authorization: Bearer <api_key>" \
  -H "Content-Type: application/json" \
  -d '{
    "system_prompt": "You are Aria, a friendly sales rep for Acme Corp. Keep replies under two sentences. When the caller says goodbye, call end_call.",
    "temperature": 0.7
  }'

Field	Default	What it does
`system_prompt`	General assistant	The agent’s persona, goals, and rules
`temperature`	`0.7`	Creativity. Lower = more consistent, higher = more varied

Switch provider or model

To use a different provider or model, pass provider and model together:

curl -X PATCH https://api-tokyo.oneinbox.ai/v1/models/<llm_id> \
  -H "Authorization: Bearer <api_key>" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "openai",
    "model": "gpt-4o-mini"
  }'

For Anthropic or Groq, pass credential_id as well:

curl -X PATCH https://api-tokyo.oneinbox.ai/v1/models/<llm_id> \
  -H "Authorization: Bearer <api_key>" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "anthropic",
    "model": "claude-haiku-4-5",
    "credential_id": "<crd_abc123>"
  }'

See Integrations for how to add a credential for Anthropic or Groq.

OpenRouter — curated model list

OpenRouter is platform-provided — no credential or integration needed. OneInbox exposes a curated subset of OpenRouter models, including small language models (SLMs) for low-latency use cases and select frontier models. Set provider to openrouter and pick a model ID from the supported list:

curl -X PATCH https://api-tokyo.oneinbox.ai/v1/models/<llm_id> \
  -H "Authorization: Bearer <api_key>" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "openrouter",
    "model": "mistralai/ministral-3b-2512"
  }'

Popular supported models include:

Model ID	Notes
`mistralai/ministral-3b-2512`	Fastest SLM — ideal for high-concurrency calls
`mistralai/ministral-8b-2512`	Balanced SLM
`mistralai/ministral-14b-2512`	Higher quality SLM
`mistralai/mistral-small-2603`	Mistral small
`openai/gpt-4o-mini`	Cost-efficient OpenAI model

The full list of accepted model IDs is available in the model dropdown on the API reference page.

Use the exact model ID — for example mistralai/ministral-3b-2512, not mistralai/ministral-3b. A wrong ID causes the call to fail silently.

Custom LLM — bring your own model server

If you run your own model (self-hosted, fine-tuned, or on private infrastructure), point OneInbox at it using a custom provider and a WebSocket or HTTP URL. Your server must expose an OpenAI-compatible streaming chat completions interface.

curl -X POST https://api-tokyo.oneinbox.ai/v1/models \
  -H "Authorization: Bearer <api_key>" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "My self-hosted LLM",
    "provider": "custom",
    "custom_websocket_url": "wss://your-llm-server.example.com/v1/chat/completions",
    "system_prompt": "You are a helpful voice assistant.",
    "temperature": 0.7
  }'

Field	What it does
`custom_websocket_url`	Your server URL — accepts `wss://`, `ws://`, `https://`, or `http://`
`provider`	Must be `"custom"`

Your endpoint receives the same request shape as OpenAI’s /v1/chat/completions and must return a streaming response in the same format. Tools, system prompts, and temperature are passed through unchanged.

Attach tools

Tools give the agent the ability to take actions during a call. They are attached to the LLM model — not to the agent directly:

curl -X PATCH https://api-tokyo.oneinbox.ai/v1/models/<llm_id> \
  -H "Authorization: Bearer <api_key>" \
  -H "Content-Type: application/json" \
  -d '{
    "tool_ids": ["<tool_id_1>", "<tool_id_2>"]
  }'

This replaces the full tool_ids list. Include all tool IDs you want active — any not listed are detached. See Tools for how to create tools.

Attach a knowledge base

Knowledge bases let the agent answer questions from your documents or URLs. They are attached to the agent — not to the LLM model directly:

curl -X PATCH https://api-tokyo.oneinbox.ai/v1/agents/<agent_id> \
  -H "Authorization: Bearer <api_key>" \
  -H "Content-Type: application/json" \
  -d '{
    "kb_ids": ["<kb_id_1>", "<kb_id_2>"]
  }'

This replaces the full kb_ids list. Include all knowledge base IDs you want active — any not listed are detached. See Knowledge bases for how to create and manage them.

Get an LLM model

curl https://api-tokyo.oneinbox.ai/v1/models/<llm_id> \
  -H "Authorization: Bearer <api_key>"

{
  "id": "llm_xyz789",
  "provider": "openai",
  "model": "gpt-4o-mini",
  "system_prompt": "You are Aria...",
  "temperature": 0.7,
  "tool_ids": ["tool_abc123"]
}

API reference

List models · Get model · Update model

Get started

Make calls

Configure your agent

Monitor & integrate

Core concepts

How LLMs work

Providers

Set a system prompt

Switch provider or model

OpenRouter — curated model list

Custom LLM — bring your own model server

Attach tools

Attach a knowledge base

Get an LLM model

API reference

​How LLMs work

​Providers

​Set a system prompt

​Switch provider or model

​OpenRouter — curated model list

​Custom LLM — bring your own model server

​Attach tools

​Attach a knowledge base

​Get an LLM model

​API reference

How LLMs work

Providers

Set a system prompt

Switch provider or model

OpenRouter — curated model list

Custom LLM — bring your own model server

Attach tools

Attach a knowledge base

Get an LLM model

API reference