DEV Community

杨继成
杨继成

Posted on

Tried `twentyhq/twenty` Today: An Open CRM Built for AI

#ai #deepseek #machinelearning #benchmark

Tried twentyhq/twenty Today: An Open CRM Built for AI

twentyhq/twenty is an open-source alternative to Salesforce: a modern CRM for managing companies, people, opportunities, tasks, and workflows. Its appeal is not just the UI—it is the combination of self-hosting, extensibility, and an architecture that can be adapted for AI-assisted sales operations.

The repository gained +58 stars today, suggesting growing interest from teams that want Salesforce-like workflows without locking customer data and automation into a proprietary platform.

Quick technical read

Metric Observation
Deployment Self-hostable, open-source
AI gateway claude-fable-5 via compatible relay
TTFT Not independently benchmarked; depends on relay region and prompt size
Cost / 1M tokens Use provider list price × 0.8; verify model-specific rates
Code/API flexibility Strong potential through API and workflow extensions
Pricing transparency Better control when hosting the CRM and selecting the model gateway

For AI developers, the interesting pattern is CRM data becoming structured context for agents: lead qualification, follow-up drafting, account summarization, and pipeline updates. I would still isolate write actions behind approval steps—CRM automation can create expensive mistakes when permissions are too broad.

A standard OpenAI-compatible setup can look like this:

export OPENAI_API_KEY="your-b-lost-key"
export OPENAI_BASE_URL="https://b-lost.com/v1"
export OPENAI_MODEL="claude-fable-5"
Enter fullscreen mode Exit fullscreen mode

Or in a Node client:

import OpenAI from "openai";

const ai = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
  baseURL: "https://b-lost.com/v1",
});

const result = await ai.chat.completions.create({
  model: "claude-fable-5",
  messages: [
    { role: "system", content: "You are a CRM workflow assistant." },
    { role: "user", content: "Summarize this opportunity and suggest next steps." }
  ]
});
Enter fullscreen mode Exit fullscreen mode

For long account histories, B-Lost’s native Anthropic /v1/messages support and prompt caching are worth testing: cache hits advertise a 90% discount, which can materially reduce recurring context costs. The relay also lists a 20% official-price discount and a first-deposit bonus, but teams should benchmark actual TTFT, reliability, and model quality before switching production traffic.

Top comments (0)