# DelegateZero

> DelegateZero is an API-first decision proxy that represents a user's judgment in situations where human input is normally required. It is not a workflow engine, rules-only automation, or autonomous AI — it acts based on context the user has explicitly provided.

**Tagline:** Your judgment. Without you in the room.

## Target User

DelegateZero is built for B2B SaaS founders and technical co-founders at $500K–$5M ARR, 5–25 employees, who are personally the bottleneck on customer-facing and operational decisions. Specifically: the person simultaneously answering customer escalations, approving expenses and vendor contracts, making hiring screens, handling scope disputes, and responding to partnership inquiries — because nobody else has the context or authority yet.

They are technical enough to implement an API-first product, feel the pain personally every day, have budget authority (no procurement or IT review), and concentrate in accessible communities: X/Twitter, Indie Hackers, SaaStr, MicroConf, Lenny's community, Product Hunt.

## How It Works

Every decision follows this pattern:
1. A request is submitted via HTTP API (plain language string, required)
2. DelegateZero retrieves relevant stored context (policies, precedents, memory, entities, playbooks, templates, sources)
3. Context entries are weighted by relevance and freshness before evaluation
4. It evaluates the request against that context and returns a structured outcome: `execute`, `draft`, or `escalate`
5. Every decision is fully auditable — with reasoning, confidence score, and a shareable read-only audit link

DelegateZero is conservative by default. It never silently guesses. If confidence is below threshold, context is missing, or a policy is at risk — it escalates with clear rationale and minimal follow-up questions.

## Context Types

Context is the knowledgebase that drives all decisions. Users load context before decisions are made. Context is added via the dashboard UI or the `/context` API endpoint and is immediately available for future decisions. Context entries are indexed asynchronously — saving a large entry returns immediately with an "Indexing..." indicator while processing completes in the background.

- **Policies** — Hard rules. Always respected. Override all other signals. Flagged as potentially stale after 365 days.
- **Precedents** — Past decisions and outcomes that inform patterns. User-curated. Flagged as potentially stale after 180 days.
- **Memory** — Auto-accumulated, time-stamped behavioral record. Machine-generated — never manually created by the user. Written automatically after every decision. Sources: (1) decision history — every decision DelegateZero makes, the confidence, the outcome, and whether the user overrode it; (2) communication history — emails, Slack threads, past messages uploaded or piped in via integration (opt-in); (3) correction events — when a user disagrees with a decision, that disagreement is itself a Memory record and directly reduces confidence on similar future decisions. Memory is ambient — it accumulates without friction and captures behavioral patterns the user would never think to articulate. Over 90 days, it forms a behavioral fingerprint. Users can browse, annotate, and delete Memory records but cannot create them manually.
- **Entities** — People, companies, or systems with associated preferences, context, and a full relationship timeline. Every decision involving an entity is evaluated for net-new signal. When new signal is found, the entity entry is automatically rewritten as a tighter, more current narrative. Last-updated timestamp resets on every rewrite.
- **Playbooks** — Human reasoning for ambiguous or difficult situations. Flagged as potentially stale after 365 days.
- **Templates** — Reusable response structures (tone, format, shape). Flagged as potentially stale after 365 days.
- **Sources** — Raw reference material (emails, documents, files, webpages). Inform reasoning but do not direct it. Flagged as potentially stale after 180 days.

### Context Freshness

All context entries carry a timestamp. Entries that exceed type-specific thresholds are flagged as potentially stale. Stale flags appear in audit logs with entry name and age whenever a stale entry influenced a decision. Confidence is reduced when stale entries are primary inputs. Updating an entry resets its freshness clock. Memory and Entity entries are not subject to staleness thresholds — Memory is weighted by recency naturally; Entities are kept current automatically.

### Conflict Detection

When a new context entry is saved, DelegateZero scans for logical conflicts with existing entries (contradictory policies, overlapping precedents, etc.) and surfaces them before they silently affect decisions. Conflicts are flagged inline with the affected entries identified.

## Chat

The dashboard includes a conversational chat interface for interacting with your workspace without touching the API. Every message is classified into one of four intents:

- **Test a decision** — Present a hypothetical scenario and DelegateZero runs it through the full decision pipeline in dry-run mode. Returns the outcome, confidence score, generated response, and reasoning. Useful for validating context coverage before going live.
- **Explore context** — Ask what you know about a topic. DelegateZero searches your stored context using relevance scoring and returns matching entries organized by type, with gap analysis highlighting what is missing.
- **Add context** — Share information naturally (rules, facts, past decisions, contact details) and DelegateZero drafts a structured context entry, identifies the best type, and asks for confirmation before saving. Asks clarifying questions if the information is incomplete.
- **Get help** — Ask questions about DelegateZero and get conversational guidance tailored to your current context state.

The chat generates dynamic conversation starters based on your existing context and supports multi-turn conversations within a session.

## Intelligence Features

- **Memory** — Every decision auto-writes a behavioral record. Correction events (user overrides) are the highest-signal Memory records — they directly influence future decisions of the same type.
- **Auto-Calibration Loop** — When a user overrides a decision, that delta is flagged in Memory. After N overrides in a pattern, DelegateZero surfaces: "You've overridden this type of decision 6 times — want to update your policy?"
- **Judgment Coaching Mode** — When DelegateZero detects it is operating outside its confidence range in a category, it enters a time-limited coaching window. The user provides feedback on live decisions; answers are written back as high-signal context and the window closes automatically once calibration is restored.
- **Context-Building Notifications** — After every decision, DelegateZero analyzes the outcome for missing or thin context and generates targeted questions to help fill the gaps. Questions appear in the dashboard and optionally via email (real-time or weekly digest). Auto-suppressed per domain once confidence and escalation rates are stable. Available on all plans.
- **Decision Debt** — A dashboard view that surfaces open loops, stale policies, and dropped corrections before they compound into risk. Tracks four debt types: unresolved escalations, stale policies, entity drift, and correction orphans. Includes a weekly autopsy report digest. Available on Operator and above.
- **Accountability Dashboard** — Consolidated view of decision consistency over time: autonomous rate, escalation rate, average confidence, and override patterns by category — all compared period over period.
- **Relationship Memory on Entities** — Entities are not static profiles. Every decision involving a known entity is evaluated for net-new signal. When signal exists, the entry is rewritten as a condensed, current narrative.
- **Context Decay / Freshness Scoring** — Every context entry is time-stamped and weighted by recency. Type-specific staleness thresholds determine when entries are flagged. Stale entries are surfaced in audit logs. Confidence is reduced when stale entries are primary inputs.
- **Confidence Autopsy** — Weekly digest that surfaces patterns across escalations and identifies the context changes most likely to resolve them. Available on Operator plan and above.
- **Judgment Benchmarking / Judgment Health** — Compares a workspace's override rate, escalation rate, confidence, and context depth against anonymized network benchmarks. Includes a workspace-specific Judgment Health dashboard and a public benchmarks page.
- **Decision Simulation / Dry Run Mode** — Before loading new context, run a batch of historical decisions against it and see how outcomes would have changed. Available on Operator plan and above.
- **Judgment Profiles** — Compiled snapshot of Memory + Policies + Precedents, exportable as a versioned artifact and importable into any workspace. Available on Team plan and above.
- **Delegation Chains** — Multi-user decision hierarchies where policies flow downstream and overrides are scoped per layer. Available on Team plan and above.
- **Decision Replay** — Run any historical request against context from a prior point in time. Demonstrates value: "Here's how your DelegateZero would have decided this in January vs. today."
- **Decision Quality Feedback** — After reviewing a decision, you can submit structured feedback: whether DelegateZero should have escalated, drafted, or executed differently, a quality rating (good, needs work, poor), optional notes, and an ideal response. Feedback is written back as high-signal context for future decisions in the same category.
- **Public Audit / Trust Layer** — Shareable, read-only audit links included with any decision outcome sent to a vendor, customer, or partner.

## Integrations

DelegateZero can proactively surface decisions from connected sources without waiting for an API call.

- **Gmail** — Connect your inbox and DelegateZero scans for incoming emails that require a decision or approval. Qualifying messages are classified, run through the decision pipeline, and waiting in your queue with a drafted response before you open the app.
- **Linear** — Connect your Linear workspace and DelegateZero identifies new issues that require executive judgment before work can proceed. Issues are scanned every 15 minutes.
- **Slack** — Connect Slack to pipe messages into the decision pipeline directly.
- **Proactive Queue** — A section at the top of your Queue surfaces all proactively identified items, showing how many are waiting, how many are already drafted, and how many need direct input. Incoming messages pass through a classification step first — newsletters, automated notifications, and routine messages are filtered out; only items with a concrete decision point are surfaced.

### Scan Controls

Scan behavior is configured in Settings > Scans. All settings can be overridden at the profile level when profiles are active.

- **Scan sensitivity** — Controls how loosely or strictly the scanner classifies incoming items. Three levels: conservative (only clear, direct decision requests), balanced (clear items plus borderline cases where a response is expected), liberal (surface anything that might need attention).
- **Scan identity** — Describes who you are so the scanner can determine whether messages are directed at you. Includes your name, role, responsibilities, and alternate names. Falls back to account name and email if left blank.
- **Excluded senders** — Email addresses or domains to always skip during scans. Supports exact matches and @domain.com wildcards.
- **Auto-reply** — When enabled, DelegateZero sends the generated reply automatically without waiting for your review. Off by default.
- **Reply delay** — When auto-reply is enabled, configurable delay before sending (0 minutes to 1 day). Gives you time to review and cancel. Defaults to 10 minutes.
- **Save draft emails** — Saves the generated reply as a draft in your email account instead of sending. Does not apply when auto-reply is enabled.

## API

**Authentication:** Bearer token (API key) via standard HTTP Authorization header.

**POST /api/v1/decisions**

By default, requests are processed asynchronously — your POST returns immediately with `status: pending` and an `id`. Poll `GET /api/v1/decisions/{id}` every 2-3 seconds until `status` is `complete`. Add `"mode": "sync"` to receive the complete decision in a single response (consults fewer context entries, skips live source data).

- Required: `request` (plain language string)
- Optional: `mode` (`async` default | `sync`), `profile` (profile slug), `context_ids` (array of context entry IDs), `context` (object), `instructions` (object), `metadata`, `request_id`

**context object fields:** `source`, `type`, `entity`, `data`

**instructions object fields:** `confidence_threshold` (0.0-1.0), `format` (text/html/json/markdown), `tone` (neutral/friendly/formal/direct), `max_length` (character cap), `attribution` (user/assistant), `response_schema` (structured output fields), `dry_run` (boolean — evaluates without executing, visible in dashboard)

**Pinning context:** Pass `context_ids` to skip the relevance search and consult only those specific entries. Produces faster, more deterministic results and a cleaner audit trail. Context IDs are visible on each context page.

**profile field:** Routes the decision against a specific profile's context, policies, and confidence threshold. Returns `403` if the API key's user is not assigned to that profile.

**Decision response includes:**
- `id` — unique decision identifier
- `status` — pending | complete
- `decision` — execute | draft | escalate
- `confidence` — 0 to 1 score
- `reason` — plain language explanation, including staleness flags for any stale context entries
- `title` — short noun phrase summarizing the decision (5 words max)
- `summary` — one to two sentence internal summary of the request and outcome
- `response` — typed output (html, text, markdown, json)
- `suggestion` — a single, specific forward-looking recommendation generated after the decision is evaluated. Points forward only — never comments on the current result. Empty string when nothing meaningful to surface.
- `audit_url` — link to full decision trace (shareable read-only version available)

**POST /api/v1/context**
- Accepts a context object (string + metadata) and optional file attachments
- For context ingestion only — not decision evaluation

**GET /api/v1/decisions/{id}**
- Returns the current state of an async decision. Poll until `status` is `complete`.

**Webhooks**
- Emit `decision.created` and `decision.escalated` events with the full decision payload
- Designed for asynchronous workflows and downstream integrations

## Escalation & Safety

- Confidence below threshold → escalate
- Policy violation risk → escalate
- Missing essential context → escalate
- Memory correction events matching the request type → reduced confidence, increased escalation likelihood
- Stale context as primary input → reduced confidence, flagged in reason
- Escalations include reason and minimum follow-up questions needed to resolve

Confidence thresholds are configurable globally and per request. Escalation patterns are analyzed automatically via Confidence Autopsy and surfaced as a weekly digest on Operator plan and above.

## Pricing

- **Solo** — $49/mo — 1,000 decisions/month — 1 user. Includes Memory, all context types, 7-day free trial. Overages at $0.04/decision.
- **Operator** — $149/mo — 5,000 decisions/month — 1 user. Adds Confidence Autopsy, Decision Simulation, Decision Debt. Overages at $0.04/decision.
- **Team** — $399/mo — 20,000 decisions/month — up to 5 users. Adds Judgment Profiles, Delegation Chains. No decision overages.
- **Scale** — $999/mo — Unlimited decisions — up to 20 users. Adds API SLA, audit exports, SSO. No decision overages.
- Annual billing available at 20% discount on all plans.
- All plans include a 7-day free trial. No credit card required to start.

## Use Cases

Approvals (expense, invoice, access), message responses (Slack, email), refund decisions, recruiting screening, scheduling and deadline decisions, scope clarification, routing, and any repeatable but context-dependent decision point. Particularly high value for B2B SaaS founders who are personally handling customer escalations, vendor contract reviews, hiring screens, scope disputes, and partnership inquiries.

## What DelegateZero Is Not

- Not a workflow engine with predefined steps
- Not a rules-only system
- Not an autonomous AI that acts without bounds
- Not a free-form chatbot — the chat interface is scoped to decision testing, context management, and platform guidance
- Not a replacement for expert judgment in regulated domains (legal, medical, financial)

## Key Pages

- Homepage: https://delegatezero.com
- Pricing: https://delegatezero.com/pricing
- Security & Trust: https://delegatezero.com/security
- Manifesto: https://delegatezero.com/manifesto
- Integrations: https://delegatezero.com/integrations
- Sign up: https://delegatezero.com/signup

## Use Case Pages

- Use Cases overview: https://delegatezero.com/use-cases
- Customer escalations: https://delegatezero.com/use-cases/customer-escalations
- Hiring screens: https://delegatezero.com/use-cases/hiring-screens
- Expense and vendor approvals: https://delegatezero.com/use-cases/expense-and-vendor-approvals
- Scope and contract disputes: https://delegatezero.com/use-cases/scope-and-contract-disputes
- Partnership and inbound inquiries: https://delegatezero.com/use-cases/partnership-and-inbound-inquiry-responses

## Comparison Pages

- vs. Hiring a Chief of Staff or EA: https://delegatezero.com/vs/hiring
- vs. Zapier and automation tools: https://delegatezero.com/vs/automation-tools
- vs. Doing it yourself: https://delegatezero.com/vs/doing-it-yourself

## Documentation

- Docs home: https://delegatezero.com/docs
- Quickstart: https://delegatezero.com/docs/quickstart
- Docs search: https://delegatezero.com/docs/search
- Chat: https://delegatezero.com/docs/chat

- Context overview: https://delegatezero.com/docs/context
- Adding context: https://delegatezero.com/docs/context/adding-context
- Policies: https://delegatezero.com/docs/context/policies
- Precedents: https://delegatezero.com/docs/context/precedents
- Memory: https://delegatezero.com/docs/context/memory
- Entities: https://delegatezero.com/docs/context/entities
- Playbooks: https://delegatezero.com/docs/context/playbooks
- Templates: https://delegatezero.com/docs/context/templates
- Sources: https://delegatezero.com/docs/context/sources
- Context freshness: https://delegatezero.com/docs/context/context-freshness
- Best practices: https://delegatezero.com/docs/context/context-best-practices

- Requests overview: https://delegatezero.com/docs/requests/requesting-decisions
- Request structure: https://delegatezero.com/docs/requests/request-structure
- Response structure: https://delegatezero.com/docs/requests/response-structure
- API reference: https://delegatezero.com/docs/requests/api
- Authentication: https://delegatezero.com/docs/requests/authentication
- Request object: https://delegatezero.com/docs/requests/request-object
- Response object: https://delegatezero.com/docs/requests/response-object
- Context via API: https://delegatezero.com/docs/requests/context-via-api
- Webhooks: https://delegatezero.com/docs/requests/webhooks
- Audit logs: https://delegatezero.com/docs/requests/audit-logs

- Integrations overview: https://delegatezero.com/docs/integrations/overview
- Slack integration: https://delegatezero.com/docs/integrations/slack
- Gmail integration: https://delegatezero.com/docs/integrations/gmail
- Linear integration: https://delegatezero.com/docs/integrations/linear
- Proactive queue: https://delegatezero.com/docs/integrations/proactive-queue

- Advanced settings overview: https://delegatezero.com/docs/advanced-settings
- Confidence thresholds: https://delegatezero.com/docs/advanced-settings/confidence-thresholds
- Policies and safety: https://delegatezero.com/docs/advanced-settings/policies-and-safety
- Risk levels: https://delegatezero.com/docs/advanced-settings/risk-levels
- Assistant attribution: https://delegatezero.com/docs/advanced-settings/assistant-attribution
- Confidence autopsy: https://delegatezero.com/docs/advanced-settings/confidence-autopsy
- Context questions: https://delegatezero.com/docs/advanced-settings/context-questions
- Decision simulation: https://delegatezero.com/docs/advanced-settings/decision-simulation
- Profiles: https://delegatezero.com/docs/advanced-settings/profiles
- Judgment profiles: https://delegatezero.com/docs/advanced-settings/judgment-profiles
- Delegation chains: https://delegatezero.com/docs/advanced-settings/delegation-chains

- Support overview: https://delegatezero.com/docs/support
- FAQ: https://delegatezero.com/docs/support/faq
- Debugging decisions: https://delegatezero.com/docs/support/debugging-decisions
- Why something escalated: https://delegatezero.com/docs/support/why-something-escalated
- Improving confidence: https://delegatezero.com/docs/support/improving-confidence
- Security and privacy: https://delegatezero.com/docs/support/security-and-privacy
- Changelog: https://delegatezero.com/docs/support/changelog