TPipe is the agent operating substrate from Ten Trillion Triangles. These are the deployments running on it and the third-party coverage written about it.
Three proof-of-concept systems running on the TPipe substrate, each documented with the architecture, the orchestration pattern, and the outcome.
The agentic step-through debugger built natively on the TPipe substrate. Manager-worker pipelines drive LLDB through the Debug Adapter Protocol, with the Pipe Context Protocol exposing debugger tools directly to the workers.
Read the case studyLong-horizon manuscript orchestration. A 15+ pipe linear refinement path with MiniBank domain isolation and ContextBank research nodes keeps a 300-page document coherent across thousands of generations.
View on GitHubA persistent game master for autonomous simulations. Swarm-to-funnel orchestration, mutex-guarded world-state writes, and headless deployment across a cluster of TPipe processes.
Read the case studyCase studies and customer stories written by Avahi and AccelByte about work they delivered with Ten Trillion Triangles. Every figure below is the publisher's own — Avahi's benchmark results and AccelByte's customer-story numbers are reported here as those publishers reported them.
Avahi documents the AWS-native benchmark platform behind TPipe's agent evaluation work: three rule-governed benchmarks, five models driven through Amazon Bedrock, and identical fixture-generated prompts on every run so the comparison stays like-for-like. The platform also ships a run-control API, an automated pipeline into curated reporting tables, QuickSight dashboards, and a trace viewer that replays any run turn by turn.
Avahi covers the AI story-writer agent built for Autogenesis: a recursive workflow that carries current game state into every generation cycle, dynamic prompt assembly that adapts as the game changes, and guardrails plus customization for tone, randomness, and narrative boundaries so players cannot exploit the narrative system. The agent is exposed to the game engine through a RESTful JSON API, with Model Context Protocol tooling for downstream integrations.
Avahi's customer story profiles TPipeWriter running on Amazon Bedrock with Amazon Nova Pro: long-horizon manuscript orchestration scored against a published rubric, with an LLM judge across eight dimensions and strict rules that block inflated scores. The same story covers the decision to buy rather than build, and the debug work that confirmed the game actually ran.
AccelByte covers the backend stack behind Autogenesis: MCP servers, the AGS CLI, and the AI plugin used by development agents to understand the backend, interact with real environments, and configure and deploy services — alongside Hermes Agent and the studio's own tooling. The Kotlin/JVM game server runs on AccelByte Multiplayer Servers, orchestrates TPipe agent turns, and sends inference requests to AWS Bedrock.
AccelByte's earlier customer story covers Revolution, the studio's competitive digital card game, and the build-versus-buy math behind its backend: authentication, player collections, currency, social systems, matchmaking, and dedicated servers taken off the internal build list so the team could ship and operate the game without standing up a backend platform in parallel.
Read the architectural deep dive, compare TPipe to the framework alternatives, or jump straight to the docs.