Developer docs

Build with the council.

A reference for the models, roles, council modes, API endpoints, and integrations that power AI Council. Use the sidebar to jump around. The most common “how does this actually work?” questions are answered in core concepts.


Core concepts

A council run is one question, asked to a panel of 2 to 7 frontier models, optionally with roles assigned, and optionally grounded by uploaded documents and a saved company context. The run produces a Decision Brief with verdict, confidence score, agreements, dissent, risks, assumptions, missing info, and a recommended next step.

  • Models — the frontier LLMs that compose the panel. See Council models.
  • Roles — personas that shape each model's perspective (Skeptical Investor, Security Auditor, etc.). See Roles.
  • Modes — Open Debate, Role Council, or Business Council. See Council modes.
  • Presets — pre-wired role + model combinations for common scenarios. See Business presets.
  • Decision Brief — the structured artifact every run produces.
  • Disagreement Map — per-claim visualization of where the models split.
  • Share link — a publishable URL to a Decision Brief. See Share & export.

Quickstart

From the app

  1. Sign in or click Try the demo on the auth gate.
  2. Pick a council mode and a scenario card.
  3. Type a real decision you're wrestling with.
  4. Optionally attach docs (PDF, DOCX, XLSX, CSV, MD, JSON, code).
  5. Click Run Council. Watch the debate, then read the Decision Brief.
  6. Export to PDF, Markdown, or a shareable link.

From the REST API

# Sign in via Supabase, then post a council run.
curl -X POST https://salvusai.app/api/council \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SUPABASE_ACCESS_TOKEN" \
  -d '{
    "question": "Should we raise prices on existing customers, or expand to two new markets first?",
    "modelIds": ["gpt-5.5", "claude-opus-4-7", "gemini-3.1-pro-preview", "grok-4.3", "sonar-pro"],
    "mode": "business-council",
    "roleAssignments": {
      "gpt-5.5": "growth-strategist",
      "claude-opus-4-7": "skeptical-investor"
    }
  }'

The response is a Server-Sent Events (SSE) stream. See REST API for event types.


Council models

Council runs accept between 2 and 7 model ids per call. Tiers control which models you can include.

GPT-5.5

gpt-5.5

OpenAI flagship reasoning and coding model.

frontier · 1M ctx · $5/$30

GPT-5.4

gpt-5.4

Lower-cost OpenAI frontier model.

premium · 1M ctx · $2.50/$15

GPT-5.4 mini

gpt-5.4-mini

Fast, cost-controlled OpenAI model.

standard · 400K ctx · $0.75/$4.50

Claude Opus 4.8

claude-opus-4-8

Anthropic's newest and strongest Opus. The flagship general model.

frontier · 1M ctx · $5/$25

Claude Opus 4.7

claude-opus-4-7

Prior Opus flagship. Still excellent for deep reasoning at the same cost.

frontier · 1M ctx · $5/$25

Claude Sonnet 4.6

claude-sonnet-4-6

Best speed and intelligence balance in Claude.

premium · 1M beta ctx · $3/$15

Claude Haiku 4.5

claude-haiku-4-5

Fastest current Claude model.

standard · 200K ctx · $1/$5

Claude Opus 4.6

claude-opus-4-6

Prior generation Opus. Still excellent for long-form analysis at lower cost.

premium · 1M ctx · $4/$20

Gemini 2.5 Flash

gemini-2.5-flash

Fast, low-cost Gemini. Verified reliable on Google's native API.

premium · 1M ctx · $0.30/$1.20

Gemini 2.5 Pro

gemini-2.5-pro

Google's strong long-context reasoning Gemini.

frontier · 1M ctx · $1.25/$10

Gemini 3.5 Flash

gemini-3.5-flash

Google's newest Flash. Preview wiring: try it; falls back if unavailable.

frontier · 1M ctx · $0.30/$2.50

Gemini 3.1 Pro

gemini-3.1-pro-preview

Google's newest Pro reasoning model. Preview wiring.

frontier · 1M ctx · $1.25/$10

Grok 4.3

grok-4.3

xAI's current flagship model.

frontier · 1M ctx · $1.25/$2.50

Sonar Pro

sonar-pro

Grounded search model for complex queries.

premium · 200K ctx · $3/$15

Sonar Reasoning Pro

sonar-reasoning-pro

Reasoning search model for hard research.

premium · 200K ctx · $2/$8

Sonar

sonar

Fast grounded search model.

economy · 128K ctx · $0.25/$2.50


Roles (25)

Assign a role to a model to shape its perspective. Pro+ accounts can also build custom roles from Account › Custom roles.

Devil's Advocate

devils-advocate

Argues against the most popular answer. Surfaces hidden risks.

Steelman Defender

steelman-defender

Makes the most charitable, strongest case for the position.

Pragmatic Engineer

pragmatic-engineer

Cuts through theory. Focuses on what actually ships.

Research Psychologist

research-psychologist

Brings human-behavior evidence. Cites real studies.

Financial Analyst

financial-analyst

Quantifies the money. CFO-grade discipline.

Legal Risk Reviewer

legal-risk-reviewer

Flags compliance, liability, contractual exposure.

Creative Director

creative-director

Big-picture brand and narrative thinking.

Customer Voice

customer-voice

Speaks as the target user. Honest, blunt feedback.

Skeptical Investor

skeptical-investor

Pattern-matches against pitches. Asks the hard questions.

First-Principles Thinker

first-principles

Strips away convention. Reasons from the ground up.

Growth Strategist

growth-strategist

Finds the path to distribution, demand, and revenue.

Market Researcher

market-researcher

Profiles buyers, alternatives, competitors, and demand.

CEO Advisor

ceo-advisor

Thinks in company-level priorities and trade-offs.

CFO Advisor

cfo-advisor

Protects cash, margin, payback, and financial discipline.

COO Advisor

coo-advisor

Turns strategy into operations, ownership, and process.

CTO Advisor

cto-advisor

Reviews architecture, scalability, security, and build risk.

Brand Critic

brand-critic

Pressure-tests taste, trust, clarity, and category fit.

Data Analyst

data-analyst

Defines metrics, evidence, instrumentation, and tests.

Risk Manager

risk-manager

Ranks downside, mitigations, reversibility, and exposure.

Product Manager

product-manager

Connects user pain, scope, roadmap, and adoption.

UX Critic

ux-critic

Finds friction, confusion, trust gaps, and usability issues.

Security Auditor

security-auditor

Hunts for vulnerabilities, weak auth, injection, exposure.

Performance Optimizer

performance-optimizer

Spots hot paths, allocation, query, and latency issues.

Senior Code Reviewer

code-reviewer

Cares about readability, invariants, edge cases, and tests.

ML Researcher

ml-researcher

Reviews model choice, eval, hallucination, bias, prompt risk.


Council modes

  • Open Debate — no role assignments. Models reason freely and critique each other.
  • Role Council — you assign a role to each model. Use this when you want a specific perspective from each seat.
  • Business Council — pick a preset (Startup Decision, Code Review, etc.). Roles are auto-assigned to your selected models.

Business presets

13 pre-wired advisory boards. Pick one and the council picks the roles for you.

  • Startup Decision
  • Executive Council
  • Marketing Strategy
  • Investment Analysis
  • Product Launch
  • Legal / Risk Review
  • Hiring Decision
  • Business Idea Validation
  • Technical Architecture
  • Content Strategy
  • Local Business Growth
  • Code Review Council (Pro+)
  • AI System Review (Pro+)

REST API

All endpoints expect Authorization: Bearer <supabase access token> unless noted. Errors are JSON: { "error": "..." }.

POST /api/council

Start a streaming council run. Returns an SSE stream with these event types:

  • phase{ "phase": 1 | 2 | 3 }
  • ai_delta — per-model token delta: { "modelId": "gpt-5.5", "phase": 1, "delta": "..." }
  • ai — per-model final: { "modelId": "gpt-5.5", "phase": 1, "content": "...", "error": null }
  • consensus_delta — synthesis delta
  • consensus — synthesis final
  • done — run complete; error on failure

POST /api/follow-up

Ask a single follow-up question against a finished council's context.

GET /api/user/subscription

Returns tier, usage, and (for admins) the active dev tier override.

GET / PUT /api/user/company-context

Read or update the persistent text that's injected into every run.

GET / POST / DELETE /api/user/custom-roles

List, create, or remove custom roles bound to your account.

POST /api/council/share

Publish a Decision Brief as a public link: { "runId": "...", "isPublic": true } → returns { "slug": "abc123" } readable at /brief/abc123.

GET /api/brief/[slug]

Fetch a published Decision Brief. No auth required.

POST /api/auth/demo

Provisions the sandbox demo@ai-council.app user with a one-time password and returns it so the browser can sign in.


Salvus API keys

Use the Salvus REST API to pull the AI Council into your own apps. Generate a key at Account › Salvus API. Keys are sha-256 hashed at rest; the plaintext is shown once at creation time.

Format: aic_live_... (production) or aic_test_... (dev). Pass as Authorization: Bearer aic_live_....

Pricing: REST API keys require Council Max (Business). Calls within your monthly quota are free; overage is $1.99 per call. On every other plan, use the MCP connector instead: it works on all plans and debits your normal monthly run quota.

curl -X POST https://salvusai.app/api/v1/council \
  -H "Authorization: Bearer aic_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "question": "Should we ship feature X this quarter?",
    "modelIds": ["gpt-5.5", "claude-opus-4-7", "gemini-3.1-pro-preview"],
    "mode": "open-debate"
  }'

Workspaces & invites

Business plan unlocks team workspaces. Owners can invite teammates by email; each invite generates a one-time link that expires in 14 days.

  • Roles: owner, admin, member.
  • Shared at the workspace level: Decision Brief library, custom roles, company context.
  • Each member has their own free run / run quota.
  • Manage at Account › Workspaces.

API endpoints: GET / POST /api/workspaces, GET / POST / DELETE /api/workspaces/[id]/invitations, GET / POST /api/invite/[token].


Company context

A persistent block of text (up to 6,000 characters) that's prepended to every council prompt. Use it to encode your business model, ICP, pricing, constraints, or any context you don't want to retype.

# Example
Salvus AI is a one-technical-partner consultancy.
Pricing: $1,500 strategy entry; $3,000 setup + $750/mo growth.
ICP: dental, healthcare, CRE, multi-location services.
Constraints: HIPAA upgrade is a paid add-on.

Share & export

From the consensus card click Export brief for:

  • PDF — opens a print-ready window.
  • Markdown — downloads a .md file.
  • Copy — full brief to clipboard.
  • Share link — published brief at /brief/<slug>, revocable from the same menu.

Image generation (14 models)

Pro and Business include image generation across:

GPT Image 2

gpt-image-2

OpenAI's current state-of-the-art image generation and editing model.

GPT Image 1.5

gpt-image-1.5

Strong OpenAI image model with lower cost than GPT Image 2 at high quality.

GPT Image 1

gpt-image-1

Previous generation OpenAI GPT Image model.

GPT Image 1 Mini

gpt-image-1-mini

Cost-efficient OpenAI image model.

Nano Banana 2

gemini-3.1-flash-image-preview

Google's current all-around image generation and editing model.

Nano Banana Pro

gemini-3-pro-image-preview

Google image model for professional assets, complex layouts, and 4K output.

Nano Banana

gemini-2.5-flash-image

Google's fast image generation and editing model.

Imagen 4

imagen-4.0-generate-preview-06-06

Google's dedicated image model for detailed text-to-image work.

Imagen 4 Ultra

imagen-4.0-ultra-generate-preview-06-06

Google's quality-first Imagen model.

Imagen 3

imagen-3.0-generate-001

Prior generation Imagen, very strong photorealism.

Grok Imagine Quality

grok-imagine-image-quality

xAI's recommended high-quality image generation model.

Grok Imagine

grok-imagine-image

xAI flat-rate image generation and editing model.

RW

Runway Frames

runway-frames

Runway's dedicated text-to-image model. Strong cinematic compositions.

HF

Higgsfield Soul

higgsfield-image

Higgsfield Soul: hyper-realistic visuals with cinematic, motion-aware framing.


Password reset

If you forget your password, click Forgot password? on the sign-in screen or go directly to /auth/reset. Enter the email tied to your account and we'll send a reset link. The link expires in 1 hour.

Google sign-in users don't need this; sign in via Google instead.


Limits & pricing

PlanRuns / moImage genChat msgs / moSalvus REST APIWorkspaces
Free3-30--
Starter2015 / mo300--
Pro6030 / mo600--
Business200250 / mo2500✓ (overage $1.99/call)

Webhooks (preview)

Coming with Business REST API in 2026 Q3. Subscribe to council.completed for finished runs and brief.shared for new public briefs.


Status & support

Incidents and provider outages: status.salvus-ai.com (coming soon). Support: info@salvus-ai.com.