Anonymous View
Skip to content

RELEASE_NOTES_v0.1.0.md

Latest commit

 

History

History
59 lines (42 loc) · 2.67 KB

RELEASE_NOTES_v0.1.0.md

File metadata and controls

59 lines (42 loc) · 2.67 KB

v0.1.0 — Initial release

tinyagent (PyPI: mozilla-ai-tinyagent) is a small, dependency-light Python implementation of the HuggingFace Tiny Agents loop, built on any-llm. It was extracted from any-agent into its own package so it can ship and version independently.

Highlights

  • One readable agent loop. TinyAgent._run_async in src/tinyagent/agent.py is the whole thing: call the LLM, dispatch tool calls (including a final_answer tool), repeat until done.
  • First-class tracing. Every run returns an AgentTrace with OpenTelemetry-compatible spans, token counts, and cost. Spans follow the GenAI semantic conventions.
  • Callbacks for everything. Subclass Callback for guardrails, metrics, observability, or intentional cancellation via AgentCancel. _TinyAgentWrapper monkey-patches call_model and per-tool call_tool so hooks fire around every LLM call and tool execution.
  • MCP-native. First-class support for stdio, SSE, and streamable HTTP servers via MCPStdio, MCPSse, MCPStreamableHttp.
  • Provider-agnostic. Built on any-llm-sdk. Swap providers by changing model_id (mistral:mistral-small-latest, openai:gpt-4o-mini, etc.).
  • Serve as a service. Run your agent over MCP or A2A with agent.serve_async(...).
  • Evaluation utilities. LlmJudge and AgentJudge for scoring runs.

Install

pip install mozilla-ai-tinyagent

Distribution name is mozilla-ai-tinyagent; import name is tinyagent.

Optional extras:

pip install 'mozilla-ai-tinyagent[a2a]'       # A2A serving
pip install 'mozilla-ai-tinyagent[composio]'  # Composio tools
pip install 'mozilla-ai-tinyagent[all]'       # everything

Requires Python 3.11–3.13.

Quickstart

from tinyagent import TinyAgent, AgentConfig
from tinyagent.tools import search_web, visit_webpage

agent = TinyAgent.create(
    AgentConfig(
        model_id="mistral:mistral-small-latest",
        instructions="Use the tools to find an answer.",
        tools=[search_web, visit_webpage],
    )
)

trace = agent.run("Which agent framework is the simplest?")
print(trace.final_output)

Relationship to any-agent

any-agent now depends on mozilla-ai-tinyagent and re-exports the shared callback / tracing types so that both packages share class identities at runtime. Existing AnyAgent.create(AgentFramework.TINYAGENT, ...) users are unaffected.

Docs