Noeta — Python SDK for durable, crash-safe AI agents

Survives crashes

State is rebuilt from an append-only event log, never held in memory. Kill a worker mid-task and another one carries on from the last step. The same log is a complete audit trail.

How recovery works

A working agent in a few lines

Install noeta-sdk, set ANTHROPIC_API_KEY, and the agent explores your project with its built-in file tools.

python

from noeta.sdk import Options, query
from noeta.sdk.providers import AnthropicProvider

result = query(
    Options(system_prompt="You are a concise coding assistant."),
    goal="What files are in this directory, and what does each one do?",
    provider=AnthropicProvider(),
    model="claude-sonnet-5",
)
print(result.answer())

The result is more than the answer: it holds every model call, tool call and token count from the run. Quickstart →

How it fits together

Every step the agent takes is appended to the event log, so any worker can rebuild the task from it. Everything the agent can do — tools, MCP, memory, the model itself — is a plugin. How it works →

Built to be extended

Everything is a pluginThe kernel ships no capabilities. File tools, web, memory, MCP, sandboxes and model adapters are all plugins that use the same public API yours does.

Any modelAnthropic, any OpenAI-compatible gateway, or the OpenAI Responses API. Switching is one line; the agent and its history stay the same.

Approval before actionRisky tool calls pause for a human. Guards can block a call before it runs; observers watch without being able to interfere.

How it compares

	Noeta	Claude Agent SDK	LangGraph	Temporal
What it is	Durable agent runtime, as a library	Agent loop library for Claude	Graph-based agent framework	Durable workflow platform
Control flow	The model decides each step	The model decides	A graph you define	Workflow code you write
What is saved	Every event; state is derived from it	The conversation	Checkpoints of graph state	Workflow history
Waiting for a human / timer	Built in, woken exactly once	Resume the conversation	Interrupt, then the caller resumes	Built in
Scaling out	Worker pool; many hosts on Postgres	One process	Up to you, or the hosted platform	A Temporal cluster
Models	Any, one line to switch	Claude	Any	—
Extra service to run	None — your process, your database	None	None	Temporal server

Pick Noeta when an agent runs unattended for a long time and you need to recover, audit and scale it. Full comparison →

Measured on public benchmarks

24/40

Terminal-Bench 2.1

40-task sample · first pass; 33/40 best of 3 attempts · noeta-sdk 0.6.28, one run · public board 58.7%–83.8% (full set)

13/15

SWE-bench Verified

15-instance subset · first pass 9/15, 4 setup timeouts re-run · noeta-sdk 0.6.10, one run

An agent built only on the public SDK (noeta-agent main, Claude Opus 4.8), run on the official harness. Each is a single run over a sample, not a full leaderboard run, and the higher figures count re-runs. Method and caveats →

NoetaAgents that keep running.

Survives crashes

Waiting costs nothing

From a script to a cluster

A working agent in a few lines

How it fits together

Built to be extended

How it compares

Measured on public benchmarks

Where to go next

NoetaAgents that keep running.

Survives crashes

Waiting costs nothing

From a script to a cluster

A working agent in a few lines ​

How it fits together ​

Built to be extended ​

How it compares ​

Measured on public benchmarks ​

Where to go next ​

A working agent in a few lines

How it fits together

Built to be extended

How it compares

Measured on public benchmarks

Where to go next