Build, run, and improve agents

Tangle is infrastructure for AI agents. You build them in the Blueprint Agent, run them in isolated sandboxes with one API key to every model, and improve them with Intelligence, which traces every run and tells you what to fix. One account, one key, one balance across all of it.

We started by building a protocol for complex services: MPC, zero-knowledge, distributed systems. That foundation maps directly to what AI agents need: isolation, coordination, and economic accountability.

Mission

Give developers the infrastructure to run AI agents in production, and a loop that makes each run better than the last.

Agents should ship with the rigor of real infrastructure: isolation by default, reviewable outcomes, and cost tied to result.

The problem

Running agents in production is brittle. They execute outside any policy, results are hard to inspect, and there is no straightforward way to see why a run failed or measure whether a change helped. Teams either build all of this themselves or avoid shipping agents at all.

What exists today

• Sandbox runtime: isolated dev containers, one per agent, pre-loaded with any coding harness. The flagship product; everything runs here.
• Inference: one OpenAI-compatible API key to every model and provider.
• Intelligence: ingests traces from sandbox runs and inference calls, and tells you why agents fail and what to change.
• Blueprint Agent: describe a project and an AI agent builds it with you in an in-browser IDE.
• Platform: one account, sk-tan- API keys, and a shared USD balance that every product bills against.

Where it goes

• Operator network: runtime hosting moves to registered operators, paid through on-chain settlement.
• Improvement delivery: today Intelligence reports what to fix; next it delivers those fixes back to the agents directly.
• Protocol-native coordination: payments, reliability, and service discovery move on-chain.

Managed cloud is live today. The decentralized operator network is the trajectory, not a current claim.

Who this is for

• Developers shipping AI agents. Run one in a sandbox.
• Teams building products with agents. Start in the Blueprint Agent.
• Operators providing compute. Operator onboarding.
• Partners evaluating the stack. Protocol overview.

The long term

As agents take on more real work, developers need infrastructure that runs them safely and a loop that improves them with evidence. Tangle is building that: build the agent, run it under policy, trace what it did, and ship the fix.