Zhipu AI Provider

Access Zhipu AI's GLM family of models through the Anthropic-compatible API. Includes GLM-5, GLM-4.7 with native chain-of-thought reasoning, and cost-effective flash variants.

Provider Name	`zhipuai`
Module	`shared.plugins.model_provider.zhipuai`
SDK	`anthropic` (Anthropic-compatible API)
Auth	API key

Highlights

GLM-5 - Flagship MoE model with agentic engineering focus
Extended thinking - Native chain-of-thought on GLM-5 and GLM-4.7
200K context - Large context windows on flagship models
Dynamic model discovery - New models appear automatically
Function calling - Full tool use support via Anthropic-compatible API

API Compatibility

Zhipu AI exposes an Anthropic-compatible API endpoint. The provider extends AnthropicProvider and inherits all message handling, streaming, and converter logic.

Quick start

from jaato import JaatoClient

client = JaatoClient(provider_name="zhipuai")
client.connect(
    project=None,
    location=None,
    model="glm-5"
)
client.configure_tools(registry)

response = client.send_message(
    "Hello from GLM!",
    on_output=on_output
)

With extended thinking

from shared import load_provider, ProviderConfig

provider = load_provider("zhipuai")
provider.initialize(ProviderConfig(
    api_key="your-key",
    extra={
        'enable_thinking': True,
        'thinking_budget': 10000
    }
))
provider.connect("glm-5")

response = provider.send_message("Complex question")
print(f"Thinking: {response.thinking}")
print(f"Answer: {response.text}")

Available Models

Flagship Models (200K context)

Model	Context	Thinking	Notes
glm-5	200K	Yes	Flagship MoE, agentic focus
glm-4.7	200K	Yes	Native chain-of-thought
glm-4.7-flash	200K	No	Fast inference
glm-4.7-flashx	200K	No	Ultra-fast inference
glm-4.6	200K	No	Previous flagship, strong coding

Balanced Models (128K context)

Model	Context	Notes
glm-4.5	128K	Balanced reasoning and coding
glm-4.5-air	128K	Lightweight, cost-effective
glm-4.5-airx	128K	Lightweight variant
glm-4.5-flash	128K	Free tier variant
glm-4.5-x	128K	Extended variant

Dynamic Discovery

The provider automatically discovers new models via Z.AI's OpenAI-compatible /models endpoint. Newly released models appear without provider updates.

List available models

provider = load_provider("zhipuai")
provider.initialize(ProviderConfig(api_key="your-key"))

# Dynamic discovery (queries Z.AI API)
models = provider.list_models()
print(models)
# ['glm-4.5', 'glm-4.5-air', 'glm-4.5-airx',
#  'glm-4.5-flash', 'glm-4.5-x', 'glm-4.6',
#  'glm-4.7', 'glm-4.7-flash', 'glm-4.7-flashx',
#  'glm-5']

# Filter by prefix
flash_models = provider.list_models(prefix="glm-4.7")
print(flash_models)
# ['glm-4.7', 'glm-4.7-flash', 'glm-4.7-flashx']

Connect to a model

# Flagship model with thinking
provider.connect("glm-5")

# Fast inference
provider.connect("glm-4.7-flash")

# Cost-effective
provider.connect("glm-4.5-flash")

Authentication

Zhipu AI uses API key authentication. Keys can be obtained from the Z.AI platform.

Get Your API Key

International: z.ai/model-api
China: open.bigmodel.cn

Credential Priority

Priority	Source
1 (highest)	`ZHIPUAI_API_KEY` environment variable
2	`api_key` in `ProviderConfig`
3	Stored credentials (`.jaato/zhipuai_auth.json`)

Credential Storage

Credentials are stored in JSON files with Unix permissions 0600:

Project: .jaato/zhipuai_auth.json (if .jaato/ exists)
User: ~/.jaato/zhipuai_auth.json (fallback)

Environment variable

# .env file
ZHIPUAI_API_KEY=your-api-key
JAATO_PROVIDER=zhipuai
ZHIPUAI_MODEL=glm-5

Interactive login (TUI)

# Login - validates and stores your API key
zhipuai-auth login

# Check status
zhipuai-auth status

# Logout - clears stored credentials
zhipuai-auth logout

Programmatic login

from shared.plugins.model_provider.zhipuai import (
    login_with_key,
    validate_api_key,
)

# Validate and store
creds = login_with_key(
    api_key="your-key",
    on_message=lambda msg: print(msg)
)

# Or just validate
valid, detail = validate_api_key("your-key")
print(f"Valid: {valid}")

Extended Thinking

GLM-5 and GLM-4.7 (non-flash variants) support native chain-of-thought reasoning. When enabled, the model shows its reasoning process before producing a final answer.

Supported Models

Model	Thinking Support
`glm-5`	Yes
`glm-4.7`	Yes
`glm-4.7-flash`	No
`glm-4.7-flashx`	No
`glm-4.6`	No
`glm-4.5*`	No

Configuration

Option	Type	Default	Description
`enable_thinking`	bool	False	Enable chain-of-thought reasoning
`thinking_budget`	int	10000	Max tokens for thinking

Enable via config

provider.initialize(ProviderConfig(
    api_key="your-key",
    extra={
        'enable_thinking': True,
        'thinking_budget': 15000
    }
))
provider.connect("glm-5")

Enable via environment

# .env file
ZHIPUAI_ENABLE_THINKING=true
ZHIPUAI_THINKING_BUDGET=15000

Access thinking in response

response = provider.send_message(
    "Analyze the trade-offs of this architecture"
)

if response.has_thinking:
    print("=== Reasoning ===")
    print(response.thinking)
    print()

print("=== Answer ===")
print(response.text)

Configuration

Environment Variables

Variable	Default	Description
`ZHIPUAI_API_KEY`	-	API key (required)
`ZHIPUAI_BASE_URL`	`https://api.z.ai/api/anthropic`	API base URL
`ZHIPUAI_MODEL`	`glm-4.7`	Default model name
`ZHIPUAI_CONTEXT_LENGTH`	Per-model	Override context window size
`ZHIPUAI_ENABLE_THINKING`	`false`	Enable extended thinking
`ZHIPUAI_THINKING_BUDGET`	`10000`	Max thinking tokens

ProviderConfig.extra Options

Key	Type	Default	Description
`base_url`	str	See env	Override API base URL
`context_length`	int	Per-model	Override context window
`enable_thinking`	bool	False	Enable chain-of-thought
`thinking_budget`	int	10000	Max thinking tokens

Base URL

The default endpoint is https://api.z.ai/api/anthropic. Do not include /v1 in the base URL — the Anthropic SDK appends it internally.

Full .env example

# .env
JAATO_PROVIDER=zhipuai
ZHIPUAI_API_KEY=your-api-key
ZHIPUAI_MODEL=glm-5
ZHIPUAI_ENABLE_THINKING=true
ZHIPUAI_THINKING_BUDGET=10000

Full programmatic config

from shared import load_provider, ProviderConfig

provider = load_provider("zhipuai")
provider.initialize(ProviderConfig(
    api_key="your-key",
    extra={
        'base_url': 'https://api.z.ai/api/anthropic',
        'context_length': 204800,
        'enable_thinking': True,
        'thinking_budget': 10000,
    }
))

provider.connect("glm-5")

provider.create_session(
    system_instruction="You are a helpful assistant.",
    tools=[...]
)

response = provider.send_message("Hello!")
print(response.text)

Function Calling

Zhipu AI supports function calling through the Anthropic-compatible API. Tool schemas are automatically converted to the correct format by the inherited Anthropic provider logic.

Supported Features

Parallel tool calls
Tool result streaming
Automatic schema conversion

Prompt Caching

Prompt caching is not supported by Zhipu AI's Anthropic-compatible endpoint. The provider automatically disables caching regardless of configuration.

Function calling example

from jaato import JaatoClient

client = JaatoClient(provider_name="zhipuai")
client.connect(None, None, "glm-5")
client.configure_tools(registry)

# Model will use tools as needed
response = client.send_message(
    "List the files in the current directory",
    on_output=on_output
)
# Model calls list_files tool

Error Handling

The provider includes Zhipu AI-specific error interpretation with actionable messages.

Exception	Cause
`ZhipuAIAPIKeyNotFoundError`	No API key in env, config, or stored credentials
`ZhipuAIConnectionError`	Auth failure (401/403) or connection issue
`RuntimeError`	Rate limit (429) or model not found (404)

Handle errors

from shared.plugins.model_provider.zhipuai import (
    ZhipuAIAPIKeyNotFoundError,
    ZhipuAIConnectionError,
)

try:
    provider.initialize(config)
    provider.send_message("Hello")
except ZhipuAIAPIKeyNotFoundError:
    print("Set ZHIPUAI_API_KEY or run: zhipuai-auth login")
except ZhipuAIConnectionError as e:
    print(f"Connection failed: {e}")
except RuntimeError as e:
    if "rate limit" in str(e).lower():
        print("Rate limited, retry later")
    elif "not found" in str(e).lower():
        print("Invalid model name")

Corporate Proxy Support

The provider supports corporate proxy environments including:

HTTP/HTTPS proxies via standard environment variables
Custom CA certificate bundles
Kerberos/SPNEGO proxy authentication

Configuration uses the same proxy settings as other jaato providers. See the Proxy & Kerberos guide for details.

Proxy configuration

# Standard proxy
export HTTPS_PROXY=http://proxy.corp.com:8080

# Corporate CA bundle
export REQUESTS_CA_BUNDLE=/path/to/ca-bundle.crt

# Kerberos proxy auth
export JAATO_KERBEROS_PROXY=true