Skip to main content
Decision Intelligence Platform
Open Source(opens in a new tab)

Stop guessing.
Simulate decisions
with AI agents.

Yukara orchestrates a council of specialized agents that debate, critique, and reason through your hardest decisions — producing structured, explainable outcomes you can actually defend.

An interactive diagram illustrating the Yukara multi-agent reasoning workflow. The diagram shows nodes for User Input, Planner (Frames the problem), three Specialists (Strategist, Quant, Risk), Critic (Debates trade-offs), and Final Synthesis, connected by animated edges showing the flow of data and debate. Below the graph, it displays: "input → planner → specialists", "debate loop active", and "synthesis · 0.94 confidence".

Used by builders & operators at

  • Lattice
  • Helios
  • Northwind
  • Quantive
  • Beacon
  • Atlas
  • Vector
  • Foundry

How it works

One question.
A council of agents.

Yukara doesn't answer — it reasons. Every decision flows through a structured pipeline of specialized agents that disagree, refine, and converge.

  1. 01
    Planner Agent

    Planner frames the problem

    A lead agent decomposes your decision into the right sub-questions and recruits the right specialists.

  2. 02
    Specialist Agents

    Specialists reason in parallel

    Domain agents — strategy, finance, risk, behavioral — each produce their independent take with sources.

  3. 03
    Debate Loop

    Critic forces a debate

    An adversarial agent challenges weak arguments. Specialists update or defend. Trade-offs surface in plain text.

  4. 04
    Final Output

    Synthesizer delivers a verdict

    A final agent produces a structured outcome — recommendation, confidence score, top risks, and what would change its mind.

Capabilities

Built for decisions
you can't afford to wing.

Six primitives that turn a single LLM call into a real reasoning system.

  • Multi-agent reasoning

    Planner, specialists, critic, synthesizer. Each agent has a job — none of them try to do all of them.

    Core engine

  • Decision simulations

    Branch your decision into 3–5 plausible futures. Compare outcomes side by side before you commit.

  • Adversarial debates

    Agents critique each other in plain English. The strongest argument wins — and you see exactly why.

  • Memory layer

    Yukara remembers your goals, constraints, and past decisions — and reasons in your context.

  • Confidence + trade-offs

    Every output ships with a calibrated confidence score and the trade-offs the agents weighted.

  • Fast by design

    Staged inference and parallel specialists keep latency low — even when the reasoning is deep.

Live preview

Watch the agents argue.
Then make the call.

Every Yukara session is a transparent debate. You see who said what, who pushed back, and how the verdict was reached.

A live preview mockup of the Yukara application interface. It displays a simulated debate session regarding a job decision. The user asks: "Should I leave my job to start the company full-time, or stay six more months and bank runway?" After a debate between the agents, the Final Synthesis concludes: "Negotiate a 3-month sabbatical first. Test conviction with limited downside. Re-evaluate at month 3 with real product signal. If traction is real, leave. If not, you keep the runway." The system reports an 82% confidence score for this verdict.

Why Yukara

ChatGPT answers.
Yukara reasons.

A general-purpose chatbot is a brilliant intern with no second opinion. Yukara is a structured council that argues until the answer holds up.

Axis
Generic AI tools
Yukara
Reasoning style

Single-perspective answer

Multi-agent debate with explicit trade-offs

Output structure

Free-form prose, hard to act on

Verdict, confidence, top risks, and what would change its mind

Memory of you

Resets every chat — no real context

Remembers your goals, constraints, past decisions

Confidence

Sounds confident regardless of evidence

Calibrated score per recommendation, surfaced upfront

Disagreement

Hidden — agrees with your last message

Adversarial critic forces the strongest counter-argument

Auditability

Black box — no traceable reasoning

Full transcript of every agent, every revision

Use cases

For the decisions
that actually matter.

  • Career

    Career moves

    Job offer vs. counter-offer. Senior IC vs. management. Yukara weighs comp, optionality, and your own goals.

    Sample prompt

    “Should I take the staff offer at the larger company or the lead role at the startup?”

  • Product

    Startup & product strategy

    Pricing, positioning, when to ship, when to kill. Specialists debate the call before you commit a quarter.

    Sample prompt

    “Do we go horizontal or stay narrow until we hit $1M ARR?”

  • Capital

    Investment decisions

    Concentrated bets, allocation shifts, exit timing. Risk and Quant agents argue the downside in plain English.

    Sample prompt

    “Trim the position now or hold through the next earnings cycle?”

  • Life

    Life decisions

    Relocation, big purchases, education, partnerships. The ones you can't Google your way out of.

    Sample prompt

    “Move cities for a 30% pay bump or stay near family and compound here?”

Open Source

Open source.
Yours to extend.

Yukara's reasoning engine is being released under a permissive license. Self-host, fork, audit, and ship your own specialist agents on top of the core. Built in the open from day one.

  • Self-hostable

    Run the entire stack on your own infrastructure. No data ever leaves your environment.

  • Bring your own models

    Plug in OpenAI, Anthropic, local Llama, or mix providers across agents in the same run.

  • Audit every token

    Every prompt, tool call, and revision is inspectable. Nothing happens behind a black box.

Star on GitHub(opens in a new tab)Read the docs(opens in a new tab)
Apache-2.0
v0.1 — public preview
A decorative mockup of a GitHub repository card for Yukara, displaying stats like 1.2k stars and 84 forks, followed by a simulated terminal window showing commands to clone, install, and run Yukara locally. The commands are: git clone github.com/himanshu-nakrani/yukara, followed by cd yukara && pnpm install, and finally pnpm run agent "ship now or wait?". The output from the agent command displays: "→ planner routing to 3 specialists", "→ strategist analyzing market timing", "→ critic stress-testing assumptions", and "✓ verdict: ship — 78% confidence".
Early access · Limited

The next big call
shouldn't be a coin flip.

Get early access to Yukara. Bring it the decisions that keep you up at night — and get back something you can actually defend.

No spam. Founders' pricing locks in for everyone on the list.