Multi-Agent Meeting Simulation
Simulate the meeting before it happens.
在 AI 多 Agent 环境里，先把会议推演一遍。
Deliberix is an open-source multi-agent meeting simulation and deliberation framework for teams, builders, and decision-makers.
让产品评审、技术争论、风险讨论和战略决策，先在 AI 多 Agent 环境里推演一遍。
Decision Readiness
84%
typed messages × 8 phases
Agents exchange typed messages across 8 structured deliberation phases — opening, exploration, challenge, synthesis, and more.
stance · role · risk prefs
Every agent carries stance, role, and risk preferences. Simulations reflect realistic human dynamics, not generic LLM responses.
consensus · risks · actions
Outputs include consensus summaries, risk flags, and action items — structured for human review and decision-making.
A control room, not another chat stream.
Live phase state, typed turns, agent stance, and decision insights stay visible in one layered workspace.
Phases
- ✓Opening
- ✓Initial Positions
- ▶Discussion
- 4Challenge
- 5Rebuttal
- 6Consensus
- 7Vote
- 8Report
Participants
- 小supportive
小张
首席产品官
- 小cautious
小李
工程负责人
- 王opposed
王经理
财务副总
- 赵neutral
赵总
法律合规顾问
初始立场：小张支持且紧迫，小李标记约 120 小时 / 2.5 人月基础设施债务，王经理预测 $4.2M 年度经常性收入风险，赵总需要 DPA 审计。公开讨论：收入潜力是否足以证明利润率压缩和实施成本的合理性？
王经理的 $4.2M 降级风险假设没有升级销售。净推荐值数据显示 67% 的中端市场客户将价格列为主要流失原因 —— 修复定价每年可通过留存单独挽回 $2.9M 年度经常性收入。净影响约为 $1.3M 年度收入阻力，而非 $4.2M。成长版升级是我们恢复利润率的途径。
你的模型假设 67% 的流失纯粹由价格驱动。我们的客户群分析显示 41% 是功能缺口驱动的 —— 客户流向集成能力更强的竞争对手。单独修复价格无法留住那 41%。我们还面临蚕食风险：如果企业采购团队要求以成长版价格重新谈判，约 $1.8M 年度经常性收入将面临风险。
Insights
Current Consensus
- 分阶段推出：新客户自 5 月 12 日起采用分层定价，现有客户享有 12 个月祖父条款。
- 入门版 25 个账户上限在人工账单期间将对账风险敞口控制在 $200K 以下。
Key Risks
- High
入门版毛利率压缩
- High
企业版年度收入蚕食
Decision Tendency
Proceed with conditions
分阶段推出满足所有四个利益相关方的约束：产品部门对 22% 中端市场流失率的紧迫性（占月度经常性收入约 38%）、工程团队约 3 个迭代期 / 约 120 小时的计量时间表、财务部门通过 25 个账户上限保证的 28% 毛利率下限，以及法务...
What teams use it for
Deliberix simulates structured deliberation before the real meeting happens — across strategy, product, engineering, and risk.
Test product ideas. Surface customer, business, design, and engineering objections. Generate product decision reports.
Simulate tech lead, backend, security, infra, product, and finance perspectives. Compare tradeoffs. Identify risks and unresolved assumptions.
Simulate product, finance, legal, sales, and customer perspectives. Surface margin, adoption, compliance, and rollout risks.
Stress-test a plan before execution. Generate mitigations and owners.
Architecture Overview
A structured loop turns messy perspectives into a traceable decision artifact, without hiding the tradeoffs.
Configure → Simulate → Report
Every Deliberix session follows a structured three-state loop. You configure the meeting — defining goal, context, agenda, and agent roster. Agents deliberate through ordered phases (Opening, Positions, Discussion, Challenge, Rebuttal, Consensus, Vote, Report). At each phase, a moderator agent enforces structure and surfaces emerging consensus, risks, and disagreements in the live Insights panel. When deliberation ends, Deliberix generates a decision-ready report with rationale, confidence score, and action items.
Technical stack
Deliberix is built on Next.js 15 App Router with React Server Components throughout. Agent orchestration runs on the Anthropic Claude API — each agent is a typed system prompt with configurable stance, goals, risk preference, and behavior tags. The simulation engine is stateless and serialisable, enabling pause, resume, and export at any phase boundary. All data stays on your infrastructure — no cloud lock-in, no vendor telemetry.
Not a chat app. A simulation control room.
这不是聊天 App，是会议仿真控制室。
Most AI tools reduce multi-agent interactions to endless chat streams. Deliberix rejects that frame. Every conversation is structured by phase.
Every message carries a type tag — argument, objection, risk, assumption, decision. These tags are not decoration; they drive the live Insights panel, which continuously surfaces consensus, disagreements, risks, and decision tendency.
A live Insights panel continuously surfaces consensus, disagreements, risks, assumptions, decision tendency, and confidence. The moderator agent enforces phase discipline and prevents the simulation from collapsing into agreement theater.
When deliberation ends, Deliberix generates a decision-ready report with transparent confidence rationale, action items, and unresolved questions — not a transcript dump.
- Phase-based structure, not endless chat
- Typed deliberation — argument, objection, risk, assumption, and more
- Decision-ready reports with transparent confidence rationale