Skip to content

TextAgents

Let agents write better agents.

Define LLM agents in text files with TOML front-matter. Your coding agents can copy examples, tweak prompts, and build evaluation harnesses alongside your application—the runtime handles all the idiomatic PydanticAI usage.

Alpha Release

This package is in early development. Expect bugs, breaking changes, and incomplete features. Use at your own risk and please report issues.

Install

uv add textagents

Quick Example

judge.txt
---
[agent]
model = "openai:gpt-4o"

[agent.output_type.reasoning]
type = "str"
description = "Step-by-step analysis."

[agent.output_type.is_valid]
description = "Whether valid."
---
Evaluate: {input}

import textagents

agent = textagents.load_agent("judge.txt")
result = agent.run_sync(input="The sky is blue.")
print(result.is_valid)  # True

Or via CLI:

uv run textagents run judge.txt --input "The sky is blue."

Next Steps