What is Tangle?

Tangle is an infrastructure company for AI agents, with a product lab and a research lab.

Tangle provides infrastructure for running AI agents, accessing models and inspecting traces. It offers hosted infrastructure; on-premises hosting is planned and is not available today.

Build with the infrastructure See the hosted assistant preview

Five layers for agent work

These layers support the products Tangle builds and the infrastructure it sells.

  1. SandboxLive service

    Isolated computers where agents can use files, a terminal, and a browser.

    Open Sandbox

  2. RuntimeCode available; no public start link

    The execution loop that runs agent work and records what happened.

    View runtime source

  3. RouterLive service

    One API for model access, provider selection, and fallback.

    Open Router

  4. IntegrationsCode available; no public start link

    Connections, grants, and approvals for the tools an agent needs.

    View integrations source

  5. ObservabilityLive service as Intelligence

    Hosted trace analysis for inspecting agent runs.

    Explore Intelligence

Evaluation is a design requirement across the stack. Tangle aims to test behavior in every layer and product. This does not mean every path has been tested. Evaluation is not a sixth infrastructure layer.

Products and research use the infrastructure

Tangle builds on the same layers it offers to developers and teams.

Product lab

Hosted assistants are the first packaged use case. The hospitality guest-request assistant is being prepared with a design partner; no live guest conversations yet. Agent Builder helps developers build domain apps. Braid runs coding agents from a terminal client. Phone lines and vertical apps are further product work.

Read about the assistant preview

Research lab

Discovery and the software factory use Tangle infrastructure to investigate how agents run, learn from evidence, and improve under review.

Explore Tangle source