Documentation

These docs are for solo builders, dev teams, and enterprise operators who want more value from every agent run. a21e acts as an Agent Performance Layer between user intent and base model execution.

The platform improves outcomes through five systems working together: intent translation, adaptive memory, secure key management, task-aware skill execution, and closed-loop learning. Start with one workflow and expand as your usage matures.

Who this is for

Different roles care about different things. Here’s what matters for each.

Why a21e

How It Works

Every request flows through a 6-stage pipeline:

Getting Started

1. Create an account at a21e.com/sign-up
2. Go to Settings and create an API key
3. Submit your first intent using the API

Submit an Intent

curl -X POST https://api.a21e.com/v1/rpc \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer a21e_YOUR_KEY" \
  -d '{
    "jsonrpc": "2.0",
    "method": "2024-01/intent.submit",
    "params": {
      "input": "Write unit tests for a React login form with email validation"
    },
    "id": 1
  }'

Execute the Package

curl -X POST https://api.a21e.com/v1/rpc \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer a21e_YOUR_KEY" \
  -d '{
    "jsonrpc": "2.0",
    "method": "2024-01/package.execute",
    "params": {
      "package_id": "PACKAGE_UUID",
      "input": "LoginForm component with email/password fields, uses React Hook Form"
    },
    "id": 2
  }'

Provider API keys (BYOK)

What is BYOK?

BYOK stands for “Bring Your Own Key.” Think of it like using your own SIM card in a phone: a21e provides the orchestration (the phone), and your API key connects directly to the AI provider (the network). The provider bills you for tokens; a21e charges one credit per enhancement (each prompt, skill, or artifact used in a run).

What you need

1. An a21e account (sign up free at a21e.com).
2. An a21e API key — go to Settings → API key.
3. An API key from your provider (see steps below for each one).

Step-by-step: OpenAI (GPT models)

1. Go to platform.openai.com/api-keys and sign in (or create a free account).
2. Click Create new secret key. Give it a name like “a21e” and click Create.
3. Copy the key immediately — you won't be able to see it again. It looks like: sk-proj-abc123...
4. If copy succeeds, you're done! Skip to “Save your key” below.

Common mistake: OpenAI keys always start with sk-. If your key doesn't start with that, you may be looking at the wrong page.

Step-by-step: Anthropic (Claude models)

1. Go to console.anthropic.com/settings/keys and sign in.
2. Click Create Key, name it, and confirm.
3. Copy the key. It starts with sk-ant-.

Step-by-step: Google (Gemini models)

1. Go to aistudio.google.com/app/apikey and sign in with Google.
2. Click Create API key. Accept Terms of Service if prompted.
3. Copy the key. It starts with AIza.

Step-by-step: xAI (Grok models)

1. Go to console.x.ai and sign in.
2. Navigate to API keys and click Create API key.
3. Copy the key and store it safely.

Save your key

Once you have your provider key, save it in a21e so it's used automatically on every request. Go to Settings → Provider keys, choose the provider, paste your key, and click Save key. Keys are encrypted with AES-256-GCM and never shown again after saving.

Three ways to use your key

Quick reference

FAQ / Troubleshooting

Adaptive memory

Adaptive memory is a unified system that learns from every interaction. Instead of losing context between runs, it auto-captures what matters and keeps memory healthy over time. It solves ten common human-AI friction points through four integrated pillars.

1. Auto-Capture Engine

Extracts knowledge, detects patterns, and tracks corrections from your interactions. Preferences, project decisions, and style choices are captured automatically — no manual tagging required. Sensitivity classification blocks security credentials, legal-privileged material, and PII from auto-capture.

2. Memory Health System

Time-based decay prevents stale memories from polluting context. A contradiction detector flags conflicting entries. Health scoring and token telemetry ensure the most relevant, healthy memories are used each run.

3. Transparency Layer

Every memory-influenced run includes attribution: which memories were used, why they were selected, and how confident the system is. Progressive disclosure shows a badge summary for quick review and full debug details when you need them.

4. Collaboration Intelligence

Decision review triggers surface choices that need human sign-off. Scope conflict resolution prevents org-level and user-level memories from contradicting each other. Hard-enforced policies cannot be overridden; soft-guidance memories can be with a logged reason.

Memory scopes

API endpoints

The memory layer is managed automatically during execution. Memory analysis, contradiction detection, and health scoring run server-side. Programmatic memory management endpoints are coming soon.

API Reference

All methods use JSON-RPC 2.0 on POST /v1/rpc. Method names use the format 2024-01/domain.action.

Intent

Package

Feedback

Analytics

Library

Prompt Lab

OpenAI Compatibility

a21e exposes an OpenAI-compatible API, so any tool that speaks the OpenAI format (Cursor, Continue, custom agents) can use a21e as a backend.

Endpoints

Model Names

If you enable Library-only mode in the app, routing and model discovery are restricted to prompts you saved in your library. Published catalog slugs are hidden from /v1/models, and execution of unsaved catalog prompts is blocked.

Cursor Setup

1. Open Cursor Settings → Models
2. Add a new model: a21e-auto
3. Set “Override OpenAI Base URL” to your a21e API URL (e.g. https://api.a21e.com/v1)
4. Paste your a21e API key as the “API Key”
5. Select a21e-auto as your active model

Authentication

API Key

All API keys use the a21e_ prefix. Pass your key via either method:

• Authorization: Bearer a21e_YOUR_KEY
• X-API-Key: a21e_YOUR_KEY

Bring Your Own Key (BYOK)

Use your own LLM provider key for execution by passing BYOK headers:

• X-A21E-Execution-Mode: byok
• X-A21E-Provider: openai | anthropic | google | xai
• X-A21E-Provider-Key: sk-...

Composite Key Format

For single-field clients (like Cursor) that only support one API key field, use the composite key format:

Example: a21e:a21e_abc123:byok:openai:sk-xyz789

The server parses this into the individual components automatically.

Error Handling

Errors follow the JSON-RPC 2.0 error format:

No match, no charge

When we can't match your intent to any prompt (error code -32002, “No matching prompts”), we never charge. Credits are only deducted when enhancements are executed (package.execute). If intent.submit returns no package, you get a clear error with fulfillment: "no_match" and credits_charged: 0 in the error details. You only pay for enhancements that deliver a matched result.

Intent Status Taxonomy

Code Examples

Handle Clarification

# If intent.submit returns status "needs_clarification":
curl -X POST https://api.a21e.com/v1/rpc \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer a21e_YOUR_KEY" \
  -d '{
    "jsonrpc": "2.0",
    "method": "2024-01/intent.clarify",
    "params": {
      "intent_id": "INTENT_UUID",
      "clarification_input": "I want Jest tests with React Testing Library"
    },
    "id": 3
  }'

Submit Feedback

curl -X POST https://api.a21e.com/v1/rpc \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer a21e_YOUR_KEY" \
  -d '{
    "jsonrpc": "2.0",
    "method": "2024-01/feedback.submit",
    "params": {
      "type": "signal",
      "execution_id": "EXECUTION_UUID",
      "outcome": "success",
      "signals": {
        "edit_distance": 0.12,
        "reprompt_count": 0,
        "test_results": { "passed": 8, "failed": 0, "total": 8 }
      }
    },
    "id": 4
  }'

Use the OpenAI Shim

# Works with any OpenAI-compatible client
curl -X POST https://api.a21e.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer a21e_YOUR_KEY" \
  -d '{
    "model": "a21e-auto",
    "messages": [
      { "role": "user", "content": "Refactor this function to use async/await" }
    ]
  }'

BYOK with Composite Key

# Single-field composite key (useful for Cursor/IDEs)
curl -X POST https://api.a21e.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer a21e:a21e_YOUR_KEY:byok:openai:sk-YOUR_OPENAI_KEY" \
  -d '{
    "model": "a21e-auto",
    "messages": [
      { "role": "user", "content": "Add error handling to this endpoint" }
    ]
  }'