Everything copadre does, in one place.

User experience

A warm welcome

Visitors are greeted with a customisable opening message.

A helpful placeholder

A customisable placeholder nudges visitors on the best way to interact with your agent.

Silent routing to the right agent

A concierge — a type of agent you build — hands each visitor to the right specialist.

Rich, readable replies

Answers come back nicely formatted — bold, lists, headings, links — not a wall of text.

Memory

The assistant keeps the full thread in mind for the length of the conversation.

No bluffing

When a visitor asks something outside your agents’ knowledge, they’re told honestly — in words you write. Nothing is made up.

App

Fully MCP compatible

Connect Claude — or any MCP-compatible AI tool — to your copadre account and build agents, test agents, review conversations, all using your favourite tool (MCP tools are made available subject to your copadre permissions).

Everything about an agent, in one place

Name it, choose its model, set its greeting and placeholder, write its instructions, and switch it on or off.

Draft, preview, publish

Edit as a draft, test it in a preview chat right in the editor, then publish when you’re ready.

Write instructions your way

Choose how you author an agent’s instructions — create them in a structured editor or just paste something you’ve prepared into the source view — whichever you prefer.

Import your agents

Create agents by importing a template from our documentation site, or refresh your agents after making changes to a local copy of your agent definitions.

Export your agents

Download any selection of your agents as a single file — tick a concierge and everything it routes to comes with it. Edit your agents however you like, including with your favourite AI tool.

Your own Claude API key

Save your key in your copadre account — we encrypt it and never show it again in full.

No passwords to remember (or leak)

Sign in with Face ID, a fingerprint or a security key — with a one-time email link as backup.

An alerts dashboard

Tabs for out-of-knowledge questions, blocks and agent issues — click straight to the conversation or agent that needs a fix, and archive what you’ve handled.

Review every conversation

Full transcripts of all your agent conversations, shown in the visitor’s own time zone.

Notifications that fit your team

Email alerts for the things that matter — completed conversations, out-of-knowledge questions, a blocked visitor, an agent that needs attention.

Broken hand-offs are flagged

If an agent tries to hand a conversation to something that doesn’t exist, copadre tells you which agent needs fixing.

Implementation

One-line install

A single line of code drops the chat onto your site — no build step, no developer. Style it with your own CSS to fit your brand.

Agent-to-agent routing

copadre’s routing mechanism lets any of your agents hand a live conversation to any of your other agents.

Concierge and specialist agents

Build a concierge — a silent router that sends each visitor to the right place. Build specialists — agents that do one job well.

Choose the model per agent

Pick the right Claude model for each agent — fast and cheap where you can, powerful where it counts.

Caching to cut your costs

Switch on prompt caching once an agent is finalised, and Claude reuses its instructions across a conversation — lowering the cost of every multi-turn chat.

Cooldown between chats

Set how long (5–60 min) a conversation should be inactive before it is classed as finished.

Branded fallback messages

Write your own wording for “message limit reached” and “beyond what I can help with”.

Set the block duration

Decide how long an abusive visitor stays blocked.

Scale from one agent to a team

Start with a single agent and grow to a routed team of specialists.

Reporting

Available through any MCP-compatible AI tool.

What people are asking

Read across conversations and surface the topics and themes coming up again and again.

This month against last month

Compare two periods and see what’s shifted — new topics appearing, old ones dropping off, questions getting more specific or more basic.

Voice of the customer

Pull the actual phrasing visitors use, not a paraphrase — useful for FAQs, product copy, or catching a feature request before it’s been asked a dozen times.

A theme report, ready to share

Turn a stack of transcripts into a short written summary for the team — the themes, the outliers, and the one conversation worth reading in full.

Daily and weekly digests

Yesterday’s conversations, or last week’s, summarised agent by agent — ready before you sit down.

Out-of-knowledge trends

Every conversation your agents couldn’t answer, over any date range you choose, so knowledge gaps don’t stay buried in the day-to-day alerts.

Conversation volume by agent

Which agents are busiest, and how that’s shifted over time — pulled straight from your conversation history.

Engagement depth

Turn counts on every conversation tell a quick answer from a long back-and-forth apart — a fast way to spot an agent that’s struggling.

Abuse and block activity

How often visitors get blocked, and whether it’s a one-off or a pattern.

A full change history, per agent

Every prompt revision, what changed, who or what made the change, and why — for any agent, any time.

Security

Protected sign-in

Rate-limited sign-ins, and single-use email links that expire quickly.

Encrypted and authenticated

Every request is encrypted in transit (HTTPS) and authenticated on every call.

Encrypted at rest

Your agent instructions, conversation contents, personal details and your Claude API key are all encrypted (AES-256-GCM).

Locked to your domains

List the websites allowed to run your agents — requests from anywhere else are turned away.

IP whitelist

Keep your office and testers from being blocked.

Multi-tier abuse handling

Turn limits pause a single conversation, not an IP address. Abusive visitors are temporarily blocked by IP.

Protected against prompt injection

Visitor messages are cleaned of any instructions meant to hijack your agent, and prompt length is restricted.

Protected against poisoned links

Every link an agent shares is checked and pointed to your own domain. A hijacked agent cannot send your visitors to an attacker’s site.

Safety

Built on Claude

Your agents run on Anthropic’s Claude models — the latest models are added automatically.

copadre guardrails on every agent

Every agent inherits copadre’s safety and quality guardrails automatically.

Quietly self-maintaining

Old conversations archive themselves after 30 days, turn pauses and abuse blocks clear on their own.

One key, one agent

An embed key only works with the agent it was issued for. It can’t be lifted and pointed at another account’s agent.

A clean slate after a block

When a block expires, the visitor starts fresh — the exchange that caused the block is never replayed to the AI.