Seine

Agentic (deep)search orchestration

A search returns an answer.
Seine returns the truth.

Scroll to follow the current

The Problem

A single AI search gives you one angle, one framing, and zero adversarial pressure. You have no way to know if the finding would survive basic scrutiny.

1 angle

0 challenges

? confidence

The Pipeline

Five depth levels. Each activates more agents, more scrutiny, more truth.

skim Route + Fuse 0 agents ~30s

scan Triage Gate 3 agents ~1-2min

dig Council R1/R2 10 agents ~3-5min

drill Research Pipeline 17 agents ~8-15min

siege Multi-round Convergence 20 agents ~20-40min

The Council

Seven cognitive perspectives. None sees another's output. All challenge the same evidence.

Synthesizer Integration + Big-picture

Contrarian Stress-test + Prove it

Lateral Hunter Adjacent domains + Analogies

Source Critic Provenance + Authority + Bias

Pattern Spotter Trends + Correlations

Blind Spot Missing perspectives

Temporal Emerging / Peaking / Declining

The (Deep)Search

Three phases. Two validation gates. A FAIL stops the pipeline.

A Discovery

hunter scout

Gate A validator

B Analysis

skeptic referee

Gate B validator

C Synthesis

adversarial confidence

The Evidence

Every claim is scored. No label is trusted without verification.

SOLID Multiple independent sources, no contradictions

0.92

SOFT Single credible source or indirect evidence

0.60

SHAKY Single biased source or conflicting evidence

0.30

UNKNOWN Insufficient evidence to assess

0.00

Confidence formula: (evidence x 0.40) + (source_quality x 0.25) + (recency x 0.20) + (agreement x 0.15)

20 agents. One truth.

Read a real dossier →

Install

Claude Code

/plugin marketplace add adambkovacs/seine-agentic-search-orchestrator-plugin

/plugin install seine

Claude Cowork

Search marketplace for seine or add by repo URL.

GitHub MIT License 20 Agents 7 Council Members Zero Dependencies No API Keys