Core Concepts

Providers

A provider is the AI runtime that powers an agent. Exponential is provider-neutral — you're not locked into one AI vendor. Different agents can use different providers, and you can switch providers per task.

Supported providers

ProviderCompanyDefault ModelBest for
ClaudeAnthropicOpusGeneral-purpose, coding, analysis, writing
GeminiGoogleGemini 2.5 ProAlternative runtime, Google ecosystem
CodexOpenAIGPT-5.4OpenAI ecosystem, specific model preferences
GLMZ.AIGLM-5.1Chinese language tasks, Z.AI ecosystem

Setting up providers

Claude is the default provider and the most deeply integrated.

  1. Go to Settings → Providers → Claude.
  2. Follow the authentication instructions.
  3. Select your preferred model (Opus, Sonnet, or Haiku).

Available models:

ModelBest for
OpusComplex tasks, deep reasoning, high-quality output
SonnetBalanced speed and quality, everyday tasks
HaikuFast tasks, simple operations, high throughput

Gemini

  1. Install the Gemini CLI on your machine.
  2. Go to Settings → Providers → Gemini.
  3. Authenticate with your Google account.

Available models: Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.5 Flash Lite.

Codex

  1. Install the Codex CLI on your machine.
  2. Go to Settings → Providers → Codex.
  3. Authenticate with your OpenAI account.

Available models: GPT-5.4, GPT-5.4 Mini, and others.

GLM

  1. Go to Settings → Providers → GLM.
  2. Enter your Z.AI API key directly in Exponential.

Available models: GLM-5.1, GLM-5, GLM-4.7, GLM-4.5 Air.

How providers work

When an agent runs, Exponential:

  1. Takes the agent's system prompt (Personality + Knowledge + Principles)
  2. Sends it to the selected provider's runtime
  3. Gives the agent access to MCP tools (comment, propose plan, etc.)
  4. Streams the agent's output back to the task activity timeline

The experience is the same regardless of provider — agents plan, you approve, they execute, you review. The provider only affects the underlying AI model's capabilities.

Choosing a provider

For most users, Claude is the recommended default. It's the most capable option and has the deepest integration with Exponential.

Consider alternatives when:

  • You have existing credits or subscriptions with another provider
  • A specific model performs better for your use case
  • You want to compare outputs across providers
  • You're working in a language or domain where another model excels

Provider per agent

Each agent can specify its own provider in the .agent file:

runtime:
  provider: gemini
  model: gemini-2.5-pro

If no provider is specified, the agent uses the system default (Claude).

Usage monitoring

Exponential tracks provider usage so you can monitor consumption:

  • Session usage — tokens used in the current 5-hour window
  • Weekly usage — tokens used over 7 days

Usage bars in Settings show your current consumption with color coding — green when healthy, red when approaching limits.

On this page