For GTM engineers
GTM engineers and AI GTM teams build and run the agents that write, sell and market. This page shows how the Amdahl learning model plugs into those agents, with a quick start for each stack: Claude Code, Claude or ChatGPT, Clay, in-house agents, campaigns and CI.
Your agents get a learning model of what wins with your buyers, through one API. GTM engineers and AI GTM teams build and run a company's agents: the AI SDR, the sequence in Clay, the Claude Code workflow, the in-house writer. Amdahl sits inside those agents. Your agent writes, and Amdahl optimizes it to win.
How it plugs in
Your agent gathers what it knows
Who the recipient is, why you write now, what you offer and the ask, from your CRM, Clay, a CSV or the agent's own research. See Context.
It calls Amdahl
POST /messages/optimize (MCP optimize_message) with that context. Add evidence: "workspace" once your data is connected, so every claim is evaluated against your customers' own conversations.
Amdahl returns the message, evaluated
One message in the sender's voice. Your code sends it, holds it, or falls back. Amdahl never sends anything itself. See Optimize before you send.
You record what happened
record_message_feedback stores what the sender did with the message. See The learning loop.
The same model serves your other agents too. Search gives a positioning, ads or content agent your buyers' own words. Research gives a call-prep agent one cited answer on an account. See Use cases.
Quick starts for your stack
| Your stack | Start here |
|---|---|
| Claude Code, Codex or Cursor | Run npx amdahl, sign in once, and do the last step it prints. See Connect your agent and the CLI. |
| The Claude or ChatGPT app | Add a connector with the URL https://app.amdahl.ai/mcp. See Claude web and desktop. |
| Clay | Send a whole Clay row as recipient.attributes (beta), next to your context. See Context. |
| An in-house agent or AI SDR | Put the optimize call between the code that knows the recipient and the send queue, with a fallback. See Optimize before you send. |
| A campaign from a CSV | One message per row, or one batch for the whole list. See Optimize a campaign. |
| A server or CI | Create a key with amdahl keys create. See API keys for servers and CI. |
| Another tool platform | See Tool platforms. |
Open-source tools
Amdahl publishes free, open-source tools and agent skills for GTM engineers at amdahl.ai/open-source and amdahl.ai/skills. The Amdahl cookbook holds copy-paste recipes and the Claude Code plugin.
Next
- What Amdahl is: the learning model in one page.
- Patterns: the call sequences that work.
- Models and limits: latency, rate limits and known limits to plan around.