Documentation

Getting Started

1. Open Studio and describe your goal

Go to Studio and type what you want to optimize: "Audit my campaigns for wasted spend", "Monitor competitor prices daily", or select a domain chip. Decisify will build an agent, generate a plan, and propose actions. No account connection required to start.

2. Connect a data source (optional)

For live data, go to Connections and link a platform via OAuth (Google Ads, Shopify, Instagram). Decisify syncs and indexes your data automatically. You can also scrape any public URL directly from Studio.

3. Review and apply

Every proposed change shows reasoning and evidence. Approve what looks right, reject what doesn't. Hit Apply and changes go live on the connected platform. Every applied run has a one-click Rollback.

Quick API call

curl -X POST https://api.decisify.ai/api/v1/runs \
  -H "Authorization: Bearer <your_token>" \
  -H "Content-Type: application/json" \
  -d '{"agent_id": "<agent_id>", "account_id": "<account_id>"}'

Get your token from Settings.

Core Concepts

Decisify is built around a few key ideas that make autonomous agents safe, transparent, and controllable.

Agents (Optimizers)

An agent is a persistent, configurable unit that analyzes data and proposes actions. Each agent has a goal, a strategy config, connected accounts, and a version history. Agents can be scheduled to run automatically or triggered on demand.

Shadow mode (default)

In shadow mode, agents read your data but never write. All runs produce action plans you review. Nothing changes until you click Apply.

Live mode

In live mode, approved changes are applied directly to the connected platform. You still review every change before applying.

Action plans

Every run produces an ActionPlan — a structured list of proposed changes with evidence, reasoning, and expected impact. You approve or reject individual actions, then apply.

Connections

Connections link Decisify to your platforms via OAuth. Currently supported: Google Ads (full read/write including MCC), Shopify (products, orders, pricing), and Instagram (analytics via Graph API). More platforms are on the roadmap.

Blueprints

Blueprints are pre-built agent templates for common use cases — waste audits, bid optimization, price monitoring, etc. Launch one from the Blueprints page and customize it for your account.

Workflows

Workflows let you chain multiple agents into a DAG (directed acyclic graph). Run agents in sequence or parallel with conditions — e.g., run a waste audit, then feed results into a bid optimizer. Trigger by schedule or webhook.

Studio

Studio is Decisify's conversational interface. Describe what you want in plain English, and the AI builds and runs agents for you — with full tool calling, scraping, and analysis capabilities.

How it works

Studio uses a streaming chat interface (SSE) powered by DeepSeek with tool calling. The agent can search the web, scrape URLs, query your connected accounts, analyze data, generate charts, and build full optimizer configs — all within the conversation.

Tool calling

During a conversation, the AI has access to tools including: web search, URL scraping (with anti-detection), Google Ads queries (GAQL), Shopify data access, file analysis, chart generation, and agent creation. Tool calls are shown in a live timeline as they execute.

Agent builder

Studio includes a guided agent builder flow. Describe your goal, answer a few clarifying questions, and Studio generates a complete agent config — strategy, thresholds, guardrails, and schedule. Review and launch directly from the chat.

Guardian hints

The Guardian intelligence system surfaces relevant insights during your conversation — patterns from past runs, common pitfalls, and optimization suggestions — shown as subtle hints in the chat stream.

Agent Levels

Decisify supports four agent levels, from simple rules to full custom code. Each level adds capability while maintaining the same safety guarantees — action plans, approvals, audit logs, and rollback.

strategy: pause_high_cpa_keywords
thresholds:
  cpa_multiplier: 1.5
  min_conversions: 10
actions:
  - type: pause_keyword
    condition: cpa > threshold.cpa_multiplier * account_avg_cpa
guardrails:
  - max_pauses_per_run: 20
  - exclude_campaigns: [branded]

Runs & Actions

A run is a single execution of an agent against your data. Every run follows a consistent lifecycle regardless of agent level.

Run lifecycle

Actions

Each run produces a set of actions — individual changes like pausing a keyword, adjusting a bid, updating a price, or flagging an anomaly. Every action includes reasoning, evidence metrics, and expected impact.

Simulation

Before applying, you can run a simulation to see projected impact. Simulations use historical data to estimate what would have happened if the proposed changes had been in place.

Rollback

Every applied run has a one-click Rollback that reverts all changes. Rollbacks are audited, so you have a complete record. Applied changes can be undone in seconds.

Artifacts

Runs can produce artifacts — charts, tables, reports, and data exports. View them in the Artifacts tab or download them. Artifacts persist across sessions.

Start a run via API

curl -X POST https://api.decisify.ai/api/v1/runs \
  -H "Authorization: Bearer <your_token>" \
  -H "Content-Type: application/json" \
  -d '{"agent_id": "agt_abc123", "account_id": "acc_xyz789"}'

Supported Domains

Decisify works across multiple business domains. Each domain has specialized tools, analysis capabilities, and action types. You can also describe any custom goal in Studio.

Policy Engine

The policy engine is Decisify's safety layer. Every proposed action passes through policy checks before being shown to you. Actions that violate policies are blocked automatically.

Default policies

• *Maximum budget increase per run: 20%
• *Maximum bid adjustment: +/- 30%
• *Minimum data threshold before destructive actions
• *Branded/critical entity protection
• *Rate limiting: max changes per run

Custom policies

Define custom policies in YAML or through the UI. Policies are versioned and auditable. They apply across all domains and agent levels.

policy: conservative_bidding
rules:
  - action: adjust_bid
    max_change_pct: 15
    require_min_clicks: 50
  - action: pause_keyword
    require_min_conversions: 20
    exclude_match_types: [exact]

Policy violations

When an action violates a policy, it's flagged with the violation reason and blocked from the action plan. View all violations in the Audit page.

Emergency freeze

Admins can trigger an emergency freeze that immediately halts all agent activity across the tenant. No runs will execute until the freeze is lifted.

Safety & Governance

Every layer of Decisify is designed to keep autonomous agents under your control.

Shadow mode

Every new agent starts in shadow mode — reads data, proposes actions, but never writes. Switch to live mode only when confident.

Approval workflows

Each run produces an action plan that requires explicit approval. Approve all, reject individual actions, or send back for revision. On Team/Enterprise plans, multi-person approval chains are available.

Spending & rate caps

Hard limits on budget increase, bid adjustment, max changes per run, and minimum data thresholds. Enforced by the policy engine — agents cannot override them.

Audit logs

Every action — proposed, approved, rejected, applied, rolled back — is recorded in an immutable audit log. Filter by agent, user, date, or action type.

Rollback

Every applied run has a one-click Rollback. Rollbacks are themselves audited. Undo mistakes in seconds.

Role-based access (RBAC)

Control who can create agents, approve runs, apply changes, and manage connections. Admins configure policies; operators run agents; viewers see results.

Guardian intelligence

The Guardian system continuously learns from agent executions. It clusters failures, generates hypotheses, runs experiments, and produces policy rules — all automatically. Insights surface in Studio as hints and feed into the policy engine.

API Reference

The Decisify API is RESTful. All endpoints require a Bearer token and are scoped to your tenant.

Base URL

Authentication

Authorization: Bearer <your_token>

Get your token via POST /auth/login or from Settings.

Example: Create an agent

{
  "name": "CPA Guardian",
  "type": "yaml_strategy",
  "account_id": "acc_xyz789",
  "config": {
    "strategy": "pause_high_cpa",
    "thresholds": {
      "cpa_multiplier": 1.5
    }
  }
}

Example: Stream a chat message

curl -N -X POST https://api.decisify.ai/api/v1/chat/stream \
  -H "Authorization: Bearer <your_token>" \
  -H "Content-Type: application/json" \
  -d '{"message": "Audit my Google Ads for wasted spend", "session_id": null}'

Returns a Server-Sent Events stream with text, tool_call, tool_result, and guardian_hint events.

Research & Academic

Decisify includes a full academic research platform with access to 200M+ papers across multiple databases. Available to all users — no account required for basic search.

Paper Search

Search across Semantic Scholar, OpenAlex, PubMed, arXiv, and CrossRef simultaneously. Results include citations, abstracts, author profiles, and institutional data. Use filters for date range, venue, open access, and minimum citations.

Research Projects

Organize papers into projects with screening workflows. Import from search results, upload BibTeX/RIS, or add by DOI. Track screening decisions (include/exclude/maybe) and export filtered sets.

AI-Powered Analysis

Generate plain-language summaries, research hypotheses, impact narratives, and field overviews using AI. Available tools: paper summarization, citation network analysis, author expertise mapping, and systematic review assistance.

Writing Assistant

Draft, improve, and polish academic prose. The writing assistant understands citation styles, academic conventions, and can rewrite for different audiences (journal submission, grant proposal, general public).

Live Data Sources

Decisify agents access 50+ live data sources in real time. Every analysis is grounded in current data, not stale snapshots.

Financial Markets

Yahoo Finance (stocks, ETFs, crypto), Alpha Vantage (technical indicators), FRED (800K+ economic series), ECB (exchange rates), Stooq (historical data). Real-time quotes and multi-year historical data.

Web & Product Data

Live scraping of 30+ retail sites including Amazon, eBay, BestBuy, Walmart, Target, Newegg, and Lowes. Anti-bot bypass stack with residential proxies, TLS fingerprint rotation, and Playwright stealth browser. Agents can monitor competitor prices, track product availability, and extract reviews.

News & Social

Google Trends (search interest over time), GDELT (global news monitoring), Google News (article aggregation), Reddit and Hacker News (community sentiment). Trending topics, breaking news detection, and sentiment analysis.

Government & Open Data

FRED (Federal Reserve economic data), NOAA (weather and climate), EIA (energy data), World Bank (development indicators), SEC EDGAR (company filings), USPTO (patent data), OpenAQ (air quality). All accessed via official APIs with automatic rate limiting.

Alternative Data

CoinGecko (crypto market data), ISS Tracker (satellite position), Disease.sh (health/epidemic data), and more. These sources enable unique cross-domain analyses that traditional tools can't provide.

Connectors

33 integrations for connecting your business tools. OAuth-based — one click to connect, automatic sync.

Advertising

Google Ads (full GAQL access, MCC support, read/write), Meta Ads (campaigns, ad sets, creative performance), Twitter Ads (campaign metrics).

E-commerce

Shopify (products, orders, inventory, pricing — read/write), WooCommerce (products, orders), Stripe (payments, subscriptions, revenue).

Analytics & CRM

Google Analytics 4 (reports, dimensions, metrics), HubSpot (contacts, deals, pipelines), Salesforce (objects, SOQL queries), Mixpanel (events, funnels), PostHog (product analytics), Intercom (conversations, users).

Databases & Storage

PostgreSQL (direct SQL), MongoDB (collections, aggregation), AWS S3 (file access), Airtable (bases, tables), Google Sheets (spreadsheet data), Google Drive (document access).

Communication & Project Management

Slack (messages, channels), Microsoft Teams (channels, messages), Discord (server data), Email (SMTP send), Jira (issues, projects), Asana (tasks, projects), Linear (issues, cycles), Notion (pages, databases), Pipedrive (deals, contacts), GitHub (repos, issues, PRs).

Billing & Usage

Simple, predictable pricing with no hidden fees. All plans include unlimited research, chat, and web search.

Plans

Free: 50 messages/day, 5 agent runs/day, 3 agents, 3 connectors. No credit card required.
Pro ($29/mo): 150 messages/day, 15 runs/day, unlimited agents, 10 connectors, API access, CSV/PDF exports.
Team ($99/mo): 500 messages/day, 50 runs/day, unlimited connectors, 5 seats, team workspaces, RBAC, audit log.

Guest Access

No signup required to try Decisify. Guest accounts get 20 messages/day and 2 agent runs/day. Data persists for 30 days. Sign up anytime to upgrade — your history transfers automatically.

Usage Tracking

Track your daily usage from the GET /api/v1/billing/daily-usage endpoint or the Studio UI. Usage resets at midnight UTC. In-progress runs always complete even if you hit the limit.

API Keys

Pro and Team plans include API access. Generate keys from Settings. API keys are per-tenant and inherit the plan's usage limits. Use the X-API-Key header for programmatic access.

Embed API

Run Decisify agents from any application. The Embed API lets you trigger runs, stream progress, and receive results via webhooks.

Start a Run

curl -X POST https://api.decisify.ai/api/v1/embed/run \
  -H "X-API-Key: <your_key>" \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "<agent_id>",
    "webhook_url": "https://your-app.com/webhook"
  }'

Returns a run_id. Progress events are sent to the webhook URL.

Stream Progress (SSE)

Subscribe to real-time run events via Server-Sent Events: GET /api/v1/embed/stream/{run_id}. Events include phase changes, tool calls, findings, and completion status.

Check Status

Poll run status: GET /api/v1/embed/status/{run_id}. Returns current phase, progress percentage, findings count, and completion state.

Authentication

Embed API uses per-tenant API keys (not JWT sessions). Keys are stored in tenant settings and validated against plan limits. Free plans do not have Embed API access — upgrade to Pro or Team.

FAQ

Frequently asked questions about Decisify.