Is this another agent framework to learn?

No. The harness owns the loop; LiteAgents translates configuration, tools, and MCP and normalizes output, with streaming types modeled after the Claude Agent SDK as plain dataclasses.

Do I reinstall everything per harness?

Install the wanted extras in one environment such as pydantic-ai and claude-sdk, compatible dependencies are reused, OpenCode also needs a global opencode-ai 1.18.29, and switching never downloads a harness.

Can I keep provider keys behind a gateway?

Yes. Set the exact alias plus api_base and api_key in model_kwargs, the alias is sent unchanged for gateway routing, and the app then uses the gateway key instead of provider keys.

What survives a crash or long run?

Durability varies by harness: DeepAgents uses LangGraph checkpoints plus a journal, others use model, tool, or provider and MCP operation replay, with start_run handles for approvals, status, and cancel.

What Python version and install steps does it need?

Python 3.11 or newer. The preview installs from a GitHub release wheel, because the PyPI name currently belongs to a different package: for example pip install 'liteagents[pydantic-ai,claude-sdk] @ '. Extras are deepagents, pydantic-ai, claude-sdk, codex, opencode-v1 and opencode-v2, and [all] installs every Python integration.

What does run() give back?

A RunResult whose text is the final answer, whose messages holds the normalized messages and whose usage preserves available native usage reports. Use run() for a final result, query() for messages as execution proceeds, and client.start_run() when you need a handle for approvals, status or cancellation.

Can I define profiles in a file instead of code?

Yes. ProfileOptions.from_yaml or from_json loads a profile from a file, expands ${ENVIRONMENT_VARIABLE} values and validates the same options as the Python constructor. Missing variables and unknown fields fail before execution rather than at runtime.

CommunityJev Agents & MCP

Liteagents

Python SDK that runs the same prompt, tools, and MCP setup across DeepAgents, Pydantic AI, Claude Agent SDK, Codex, and OpenCode by switching harness.

See highlights

Open github.com

Stars: 37
Language: Python
License: MIT
Published by: Community
Author: BerriAI
Last verified: Oct 2026

BerriAI/liteagents

Organization repository on GitHub

View on GitHub

A provider-independent agent SDK with the same query() interface as the Claude Agent SDK, allowing you to use the right model for every turn.

Forks: 2
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

Switch between DeepAgents, Pydantic AI, Claude Agent SDK, Codex, and OpenCode through one Python SDK. Pick a harness and model, run your agent, then change harness to try another framework with the same application code. Read our blog post · Website · Getting started. Install LiteAgents and the two harnesses used below (Python 3.11+): This preview uses a GitHub release wheel; the PyPI name currently belongs to a different package. No checkout, gateway, or Temporal service is needed.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import

Works forJev for Classification Jev for Routing Jev for Tool calling

Used inJev for Software & DevTools

Languages: Python

Highlights

Change profile.harness to compare frameworks with identical app code; each run starts an independent task.
Typed Python functions become harness tools, MCP servers live in profile.mcp_servers, and results normalize to text.
Same model string works across OpenAI, Anthropic, OpenRouter, Gemini, Groq, Mistral, DeepSeek, Together, and xAI.
Conversation clients share history via query; YAML or JSON profiles expand environment variables and validate early.

Watch out

Switching does not move a native session between frameworks, models must support the requested tools and settings, and the preview installs from a release wheel because the PyPI name differs. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

Tiermem

Tiered agent memory keeping compact summaries with source links over searchable raw history, where a router decides per query whether notes suffice or raw pages must be recovered.