Skip to main content
Back to News
Mianbi Intelligence and research partners release PilotDeck 2.0 agent system
Product
2 min read
CN

Mianbi Intelligence and research partners release PilotDeck 2.0 agent system

The AMW Read

The release incrementally advances agent orchestration through editable memory and model routing, but its reported cost advantage remains specific to the team's tests.
NoveltySignificance
AI Agents · Player Map
面壁智能
面壁智能

Foundation Models / LLMs

View Company Profile

Mianbi Intelligence and research partners release PilotDeck 2.0 agent system

Tsinghua University's THUNLP laboratory, Mianbi Intelligence and OpenBMB have released PilotDeck 2.0, an open-source agent operating system. It separates project files, memories and skills into workspaces, exposes memories for inspection and editing, and supports background task execution. An idle-time memory consolidation mode includes rollback. The system supports MCP, web and CLI interfaces, and messaging integrations, with deployment options for macOS, Linux, Windows and Docker under AGPL-3.0.

The release puts agent orchestration and context management at the center of product differentiation. Its routing system assigns planning to stronger models and supporting work to cheaper ones, making task economics depend partly on how models are coordinated. Across seven complex tasks, the research team reported a score of 70.6 at $3.15 using Claude Sonnet 4.6 with MiniMax-M2.7 subagents, compared with 69.1 at $18.36 for standalone Sonnet 4.6—about 83% lower cost. Those results support a narrower claim about the tested configuration; they do not establish equivalent savings across production workloads.

For builders, the concrete opportunity is to evaluate model routing alongside inspectable memory and persistent execution as one workflow system. A useful deployment trial would compare total task cost, completion quality and recovery from incorrect memories against an existing single-model setup. Workspace isolation and editable memory address operational control, while background execution raises the practical question of how much work can finish without intervention. Investors should distinguish the reported benchmark improvement from evidence of sustained adoption or a defensible commercial position.

#AIAgents #PilotDeck #OpenSource #ModelRouting #AgentMemory

#Mianbi Intelligence#PilotDeck 2.0#AI agent orchestration#model routing#related:清华大学THUNLP实验室#related:OpenBMB

How This Connects

Based on AI Agents · Player Map

  1. 1w agoMianbi Intelligence and research partners release PilotDeck 2.0 agent system · THIS ARTICLE
  2. 3w agoEveryone’s AI brings Korea’s telecom and messaging platforms into consumer agentsEveryone’s AI
  3. 3w agoManus launches Manus 2.0 and Cue personal-agent appManus
  4. 3w agoInstinct reportedly raises $1B Series C at $10B valuationInstinct
  5. 3w agoNvidia Expands OpenShell and Introduces Sentry for AI Agent SecurityNvidia
  6. 1mo agoMCP's largest spec revision since launch ships with a prompt-injection flaw that can leak credentials.Model Context Protocol (MCP) spec update

Related News

More news from 面壁智能

Stay updated with the latest news and announcements from 面壁智能.

View all 面壁智能 news

Discover AI Startups

Explore 5,000+ AI companies with VC-grade analysis, funding data, and investment insights.

Explore Dashboard