OpenViSTAI · Agents · Blockchain · Crypto
Menu
Agent Infrastructure

Cayu

Cayu is an open-source Python framework and runtime for building and running domain-specific long-horizon agents. Developers assemble an agent harness from components such as model providers, context, tools, memory, policies and verification; its runtime supports durable sessions, recorded progress, resumable workflows, approvals and recovery.

Independent OpenViST project profile, based on linked sources.

Cayu — OpenViST project profile in Agent Infrastructure.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Cayu when a Python agent must handle work across multiple tool calls or human decisions and needs persisted progress, controlled authority or recovery. Begin with the generated project for its documented credential-free checks, then configure a provider for live runs.

Who it is for

  • Python developers building domain-specific agents that need to persist state or resume work after interruption.
  • Teams that need to govern tool authority, include human approvals, and inspect agent execution as well as final output.

Documented uses

  • Compose an agent harness from model, context, tool, memory, policy and domain components.
  • Run work through durable sessions, task dispatch, resumable steps, approvals and recovery.
  • Test and evaluate an agent using included project checks and evaluation scaffolding.
Before you choose

Requirements & limits

Cayu is open source under the Apache-2.0 license. The documented quickstart requires Python 3.11 or newer and uv. Cayu Cloud is optional.

  • The generated project and its initial checks can be used without provider credentials; live model runs require a configured provider and its credentials.
  • Applications must supply their own user interface, authentication, business authorization, domain rules and completion criteria.
  • The framework supports multiple provider families, including OpenAI API, Anthropic, Bedrock, Vertex and OpenAI-compatible APIs; OpenAI subscription login is labeled experimental.
  • Native Python tools are trusted application code; configured isolation and external receipts determine guarantees at the application boundary.
  • The documented coding starter can prepare changes for delivery but does not commit, push, open pull requests, run CI or merge.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Create and verify a Cayu starter project

Generate a project and run its credential-free inspection, checks, tests and evaluation.

Before you start

  • Python 3.11 or newer
  • uv

Steps

  1. Run uvx cayu new myagent and change into the generated project directory.

  2. Run the command below to install the development dependencies.

    Documented command
    uv sync --extra dev
  3. Run the anatomy guide, inspection, strict checks, tests and evaluation with the documented the command below commands.

    Documented command
    uv run --no-sync

Expected outcome

A generated starter project is inspected and its included tests and evaluation are run without live provider credentials.

Guide limitations

  • These checks establish starter behavior, not live-model quality.
  • A live model run requires provider configuration and the credentials specified by that provider's guide.
Source-based practical guide

Configure and make a first live agent run

Run the generated agent after configuring a supported model provider.

Before you start

  • A Cayu starter project created with the documented quickstart
  • A supported provider, its model, and the required credentials

Steps

  1. Choose a supported provider and model in configuration/settings.py or through the generated CAYU_PROVIDER setting.

  2. Set the credentials specified in the provider guide.

  3. Run the documented the command below command with a support-request message.

    Documented command
    uv run --no-sync python run.py --message
  4. Inspect the emitted events and final outcome.

Expected outcome

The application makes a live run and exposes its events and final outcome for inspection.

Guide limitations

  • Provider authentication failures are configuration problems; a run that fails or pauses is a runtime outcome to assess from its recorded evidence.
Evidence and transparency

Sources & review

Source record details
Source availability
Active
Sources checked
4 October 2026
Original website title
cayu-dev/cayu · Official homepage title · retained in its source language