Auto-Optimize your agent
The framework for building and auto-optimizing the agent harness and context — hit production accuracy in minutes, not months.
Install the SDK
pip install afnio
Why Afnio
Ship 30% more accurate agents
from day one
Transparent op graph
Define your agent as ops with explicit context messages I/O. No opaque rewriting, no abstraction hiding the LLM calls. Every optimization change is a Python diff you can inspect and version-control.
123456789101112131415161718191234567891011121314Structural credit assignment
Afnio backpropagates textual gradients through the computation graph via the chain rule, delivering targeted feedback to each Parameter based on its specific contribution to the output.
Modular context
The context aggregates multiple Parameters — you set which to optimize and which to freeze. A shared Parameter accumulates feedback from every op that uses it.
Batching as a primitive
Afnio Variables carry a batch dimension as tensors do. Ops broadcast over it through a simple sync interface, dispatching concurrent LLM calls and aggregating gradients automatically — no example lists, async plumbing, or retry loops.
123456789101112131415161718191234567891011121314Examples
Built with Afnio
Build self-improving browser agents
Browser agents that learn to solve new challenges from each new browser session.
Learn more →Align complex evaluation harness
Design multi-step evaluation pipelines, score agent outputs against ground truth, and auto-optimize to beat your accuracy targets.
Learn more →Optimize classifier accuracy
Auto-optimize prompts using textual gradients. Reach production-grade precision without manual prompt engineering.
Learn more →Pricing
Start free,
scale when you’re ready
Free for solo developers, paid plans for teams in production.
See full pricing →For individuals getting started. Perfect for auto-optimizing simple agents with a small set of examples and monthly runs — no credit card required.
- ✓Up to 1 Seats
- ✓20 Private Runs/month
- ✓3 h/month Private optimize time
For teams with extended needs. Larger run quotas for auto-optimizing complex agents with more examples or weekly runs. Ship production agents that improve over time.
- ✓Up to 5 Seats
- ✓80 Private Runs/month
- ✓20 h/month Private optimize time* Additional hours billed at $1/hour
Ship your first auto-optimized agent in minutes
Free to start. No credit card required. Build optimized agents in minutes.
Install the SDK
pip install afnio