Heabsy and FlexAI join Agno as model providers

Agno's Heabsy and FlexAI classes both extend OpenAILike, so tools, structured output and streaming behave the way they do with any OpenAI-compatible provider. Each one reads its key from its own environment variable and ships with a default model:

from agno.agent import Agent
from agno.models.flexai import FlexAI
from agno.models.heabsy import Heabsy
 
heabsy_agent = Agent(model=Heabsy(id="qwen38"), markdown=True)
flexai_agent = Agent(model=FlexAI(id="DeepSeek-V4-Flash-0731"), markdown=True)
 
# Model strings resolve to the same classes
heabsy_agent = Agent(model="heabsy:qwen38", markdown=True)
flexai_agent = Agent(model="flexai:DeepSeek-V4-Flash-0731", markdown=True)
 
flexai_agent.print_response("Explain quantum computing in simple terms")

With no key set, both classes stop before the first request. Real output from resolving the two model strings:

heabsy:qwen38 -> Heabsy qwen38 https://api.heabsy.com/v1
   ModelAuthenticationError HEABSY_API_KEY not set. Please set the HEABSY_API_KEY environment variable.
flexai:DeepSeek-V4-Flash-0731 -> FlexAI DeepSeek-V4-Flash-0731 https://api.flex.ai/v1
   ModelAuthenticationError FLEXAI_API_KEY not set. Please set the FLEXAI_API_KEY environment variable.

Heabsy's qwen38 is Qwen3.8 27B with a 262,144-token context, vision input and tool calling. You turn its reasoning on or off per request with extra_body={"chat_template_kwargs": {"enable_thinking": False}}. Heabsy serves no embeddings, so knowledge bases need a different embedder.

FlexAI hosts open-weight families such as DeepSeek, Qwen, GLM, Gemma, gpt-oss, Llama and MiniMax. Its default DeepSeek-V4-Flash-0731 has a 1M context and supports tool calls. Both providers list their current models at GET /v1/models. FlexAI's chat endpoint rejects n above 1 and the verbosity parameter.

See the Heabsy cookbook and the FlexAI cookbook, and learn more about OpenAI-compatible providers and model strings in the documentation.