diff --git a/README.md b/README.md
index d18d0bd..978ba9a 100644
--- a/README.md
+++ b/README.md
@@ -1,6 +1,6 @@
 # Meister-Eder
 
-AI-powered conversational registration system for **Spielgruppe Pumuckl**, a playgroup run by Familienverein Fällanden (Fällanden, Switzerland).
+AI-powered conversational registration agent for **Spielgruppe Pumuckl** (Familienverein Fällanden, Switzerland). Parents register their child and ask questions via email — the agent handles the conversation, validates all required fields, and notifies the playgroup admin on completion.
 
 Replaces a static Google Forms workflow with an AI agent that guides parents through child registration via natural conversation — over email or a web chat interface.
 
@@ -19,20 +19,102 @@ Replaces a static Google Forms workflow with an AI agent that guides parents thr
 | Web chat | Real-time, session-based |
 | Email | Async, thread-tracked; reminders on days 3, 10, 25 |
 
-## Status
+## Prerequisites
 
-Greenfield — specification complete, implementation not yet started.
+- Python 3.13+
+- [uv](https://docs.astral.sh/uv/) (dependency manager)
 
-## Running the Email Agent
+## Installation
 
-The email agent is a plain script invoked periodically via cron — no long-running daemon needed.
-
-### Scheduling with cron
-
-```cron
-*/5 * * * * flock -n /tmp/meister-eder-email.lock python /path/to/check_email.py
+```bash
+git clone https://github.com/gurix/Meister-Eder.git
+cd Meister-Eder
+uv sync
 ```
 
-`flock -n` acquires an exclusive lock before running the script. If a previous run is still in progress when the next cron tick fires, the new invocation exits immediately (non-blocking). The lock is released automatically by the kernel when the process ends — even on crash — so stuck locks are not a concern.
+## Configuration
 
-Adjust `*/5` to whatever polling interval makes sense (e.g. `*/2` for every 2 minutes).
\ No newline at end of file
+Copy the example env file and fill in your values:
+
+```bash
+cp .env.example .env
+```
+
+### Required variables
+
+| Variable | Description |
+|---|---|
+| `AI_MODEL` | litellm model string, e.g. `anthropic/claude-opus-4-6` or `openai/gpt-4o` |
+| `ANTHROPIC_API_KEY` | API key for Anthropic models |
+| `OPENAI_API_KEY` | API key for OpenAI models (if using OpenAI) |
+| `IMAP_HOST` | IMAP server hostname for receiving parent emails |
+| `IMAP_USERNAME` | Email account username |
+| `IMAP_PASSWORD` | Email account password |
+| `SMTP_HOST` | SMTP server hostname for sending replies |
+| `REGISTRATION_EMAIL` | Sender address shown to parents |
+
+### Optional variables
+
+| Variable | Default | Description |
+|---|---|---|
+| `IMAP_PORT` | `993` | IMAP port |
+| `IMAP_USE_SSL` | `true` | Use SSL for IMAP |
+| `SMTP_PORT` | `587` | SMTP port |
+| `SMTP_USE_TLS` | `true` | Use STARTTLS for SMTP |
+| `DATA_DIR` | `data/` | Directory for conversation state and completed registrations |
+| `KNOWLEDGE_BASE_DIR` | `openspec/…/knowledge-base` | Path to admin-editable knowledge base markdown files |
+| `POLL_INTERVAL` | `60` | Seconds between inbox polls (only used when running as a daemon) |
+
+### Switching AI providers
+
+`AI_MODEL` uses [litellm](https://docs.litellm.ai/docs/providers) model strings — any supported provider works without code changes:
+
+```bash
+# Anthropic (default)
+AI_MODEL=anthropic/claude-opus-4-6
+ANTHROPIC_API_KEY=sk-ant-...
+
+# OpenAI
+AI_MODEL=openai/gpt-4o
+OPENAI_API_KEY=sk-...
+
+# Google Gemini
+AI_MODEL=gemini/gemini-2.0-flash
+GEMINI_API_KEY=...
+```
+
+## Running
+
+### As a cron job (recommended)
+
+The agent is a plain script — no long-running daemon needed. Schedule it with cron and use `flock` to prevent overlapping runs:
+
+```cron
+*/5 * * * * flock -n /tmp/meister-eder-email.lock uv run python main.py
+```
+
+`flock -n` exits immediately if a previous run is still in progress, so the script is always safe to schedule aggressively.
+
+### Manually
+
+```bash
+uv run python main.py
+```
+
+## Development
+
+### Running tests
+
+```bash
+uv run pytest
+```
+
+All tests are unit tests — no network access or API keys required.
+
+### Knowledge base
+
+The agent answers parent questions from markdown files in the knowledge base directory. These files are designed to be edited directly by playgroup admins — no code changes needed to update fees, schedules, or policies.
+
+### Adding a new AI provider
+
+Set `AI_MODEL` to any [litellm-supported model string](https://docs.litellm.ai/docs/providers) and set the corresponding API key environment variable. No code changes required.
diff --git a/src/agent/prompts.py b/src/agent/prompts.py
index e834379..79f5003 100644
--- a/src/agent/prompts.py
+++ b/src/agent/prompts.py
@@ -13,7 +13,7 @@ STEP_DESCRIPTIONS = {
     "child_name": "Ask for the child's full name.",
     "child_dob": (
         "Ask for the child's date of birth. "
-        "Validate age: indoor requires ≥2.5 years, outdoor requires ≥3 years."
+        "Validate age: indoor requires ≥2 years, outdoor requires ≥2.5 years."
     ),
     "playgroup_selection": (
         "Explain both playgroup options and ask which the parent wants "
@@ -152,7 +152,7 @@ def _build_registration_prompt(kb: KnowledgeBase, state: ConversationState) -> s
 ## Registration Flow (8 steps)
 1. greeting — greet and detect intent
 2. child_name — ask for child's full name
-3. child_dob — ask for date of birth; validate age (indoor ≥2.5 yrs, outdoor ≥3 yrs)
+3. child_dob — ask for date of birth; validate age (indoor ≥2 yrs, outdoor ≥2.5 yrs)
 4. playgroup_selection — present options, collect type(s) and day(s)
 5. special_needs — ask about special needs / allergies / medical conditions
 6. parent_contact — name, street address, postal code, city, phone, email