Threadplane

Threadplane is a production-ready agent UI framework for Angular, published under the MIT license by Brian Love (d/b/a cacheplane). It provides the Angular UI layer that agent runtimes like LangGraph, AG-UI, CrewAI, Mastra, and Pydantic AI assume teams have already built, covering durable threads, human-in-the-loop interrupts, subagents, planning, memory, and generative UI. The core libraries are available on npm and GitHub, with a commercial license required for the @threadplane/chat drop-in component in production for-profit contexts.

What It Is

Threadplane is an Angular-native SDK for wiring AI agent backends into production Angular applications. Rather than replacing backend runtimes, it solves the frontend surface problem: how to connect a streaming, stateful agent to Angular's dependency injection, Signals API, and RxJS ecosystem without a React rewrite. The framework exposes provideAgent() and injectAgent() as the primary API — analogous to LangGraph's React useStream() hook — and returns signal-driven access to messages, status, tool calls, interrupts, subagents, history, and thread management with no subscriptions, no async pipe, and no zone.js required.

Package Architecture

Threadplane ships as a suite of focused npm packages, each with its own license:

• @threadplane/langgraph — LangGraph Platform adapter; MIT-licensed
• @threadplane/ag-ui — AG-UI protocol adapter for CrewAI, Mastra, Pydantic AI, Microsoft Agent Framework, AWS Strands, and CopilotKit Runtime; MIT-licensed
• @threadplane/chat — Drop-in chat UI (headless primitives, opinionated compositions, interrupts, generative UI); dual-licensed PolyForm Noncommercial + Threadplane Commercial
• @threadplane/render — JSON-spec-to-Angular-component engine powering generative UI; MIT-licensed
• @threadplane/a2ui — A2UI protocol types, streaming parser, and dynamic-value resolver; MIT-licensed, no Angular dependency
• @threadplane/licensing and @threadplane/telemetry — Supporting utilities; MIT-licensed

The runtime-neutral Agent contract is the stability boundary between adapters and the chat surface, so swapping @threadplane/langgraph for @threadplane/ag-ui requires no changes to chat components or templates.

Core Capabilities

• Durable threads — persist across reloads, resume, branch, and replay via threadId signal and durable transports
• Resumable interrupts — human-in-the-loop pause/resume with interrupt() and resume(), wired into LangGraph interrupts natively
• Tool calls as events — stream progress, structured args, and surfaced errors via toolCalls() on injectAgent()
• Streaming state as signals — messages(), status(), and error() are Angular Signals, not promises
• Generative UI — Vercel json-render and Google A2UI both supported, rendered into existing Angular component libraries with per-component fallback and a readiness gate
• Recoverable errors — retry, reload, error boundaries, and fallback content via error(), reload() signals
• Observability hooks — tracing seams with app telemetry off by default; opt-in only

Model and Runtime Compatibility

Threadplane is explicitly backend-agnostic on both the model and runtime dimensions. Model providers listed on the homepage include OpenAI, Anthropic, Google Gemini, Azure OpenAI, and Amazon Bedrock. Agent runtimes supported include LangGraph (native adapter), and via the AG-UI protocol: CrewAI, Mastra, Pydantic AI, Microsoft Agent Framework, AWS Strands, and CopilotKit Runtime. The Angular surface layer remains unchanged when switching runtimes.

Update: v0.0.52

The GitHub repository shows the latest release as v0.0.52, published on June 20, 2026, with the repository last pushed the same day. All packages follow a patch-only 0.0.x release policy, ensuring no minor or major version bumps silently break lockfiles. The project currently supports Angular 20 and 21 (tested in CI), with Angular 22 listed as planned and Angular ≤19 unsupported. The repository had 99 stars and 11 forks at the time of data collection, per the GitHub project page.

Pilot-to-Prod Delivery Program

Beyond the open-source libraries, Threadplane offers a concierge delivery engagement called Pilot-to-Prod — an 8-week program structured around four phases: Discover (stack mapping), Build (working demo on real data), Harden (observability, error boundaries, deploy paths), and Train (team ownership with a runbook). The program is positioned as optional; the MIT libraries are complete standalone. Enterprise licensing adds volume coverage, SLA, security review, Slack Connect support, and procurement support including MSA and custom indemnification.