analogaisdk 0.5.0

Minimal Python SDK for AnalogAI completion endpoints

AnalogAI SDK (minimal)

Tiny Python client for the Deepthink OpenAI-compatible chat server (analogai.deepthink.chat). It calls POST {CHAT_BASE_URL}/v1/chat/completions — a stateful endpoint: memory tenancy is keyed by user_id + agent_id (both created server-side on first use), Deepthink memory is retrieved per turn, and messages are ingested in the background.

Install

pip install analogaisdk

Setup

Set environment variables in .env:

• CHAT_BASE_URL (default https://cloud.analogai.net:8010; set to http://localhost:8010 for a local server)
• ANALOGAI_API_KEY (optional, if your server requires auth)
• ANALOGAI_MODEL (optional; server default is used when omitted)

Usage

from analogaisdk import AnalogAIClient

client = AnalogAIClient(agent_id="my-agent", user_id="my-user")

# Non-streaming (default).
print(client.generate_completion("are humans mortal?"))

# Streaming.
for chunk in client.generate_streaming_completion("are humans mortal?"):
    print(chunk, end="", flush=True)

Working memory

Working memory (a string or list of strings) is appended to the LLM system prompt alongside the Deepthink memory:

client.generate_completion(
    "is Socrates mortal?",
    working_memory=["Socrates is a human"],
)

Explicit messages and model

client.generate_completion(
    messages=[
        {"role": "user", "content": "Say hello"},
    ],
    model="azure:gpt-4o-mini",  # optional, provider-prefixed
    temperature=0.2,
)

Token usage and per-request USD cost

Every call surfaces the gateway's usage block (same field names as the OpenAI API: prompt_tokens, completion_tokens, total_tokens) plus a USD cost computed from a pricing table.

Built-in defaults price gpt-5.4-mini at $0.75 per 1M prompt tokens and $4.50 per 1M completion tokens. Override or extend via the pricing= constructor argument (values are USD per 1,000,000 tokens):

client = AnalogAIClient(
    agent_id="my-agent",
    user_id="my-user",
    pricing={
        "gpt-5.4-mini": {"prompt": 0.75, "completion": 4.50},
        "azure:gpt-4o-mini": {"prompt": 0.15, "completion": 0.60},
    },
    print_usage=True,  # also prints a stderr summary per request
)

Non-streaming — the return value is still a str, plus usage/cost:

result = client.generate_completion("are humans mortal?")
print(result)                    # assistant text
print(result.prompt_tokens)      # int
print(result.completion_tokens)  # int
print(result.total_tokens)       # int
print(result.cost_usd)           # float, USD
print(result.usage.as_dict())    # full breakdown incl. prompt/completion cost

Streaming — .usage is populated once the iterator is drained (the SDK sends stream_options={"include_usage": True} so OpenAI-compatible gateways emit a final chunk carrying the usage block):

stream = client.generate_streaming_completion("are humans mortal?")
for chunk in stream:
    print(chunk, end="", flush=True)
print(stream.usage.prompt_tokens, stream.usage.completion_tokens, stream.usage.total_cost_usd)

The most recent usage is also cached on the client as client.last_usage. Set ANALOGAISDK_PRINT_USAGE=1 in the environment (or pass print_usage=True) to print a one-line summary to stderr after every request:

[analogaisdk] model=gpt-5.4-mini prompt_tokens=812 completion_tokens=134 total_tokens=946 cost_usd=0.001212