Source prospects from a single sentence

Sinky (formerly CopilotCRM.ai) is a multichannel prospecting platform that combines lead sourcing, AI messaging, and campaign automation. Historical sourcing relied on structured criteria: CSV imports, LinkedIn/Sales Navigator URLs, multi-layer filters. The team wanted to raise the bar. Let a salesperson describe the need the way they would say it out loud, and let the AI find the right prospects.

Turn a free-form sentence ("IT directors in agri-food, 50 to 100 employees, in France, who raised funds in the last 5 years") into an actionable search across heterogeneous sourcing APIs. Understand implicit filters (industry, size, geography, growth signals), pick the right tool for each criterion, orchestrate the calls, deduplicate, qualify. All in a few seconds, with production-grade precision.

AI agent built on OpenAI Agents SDK in TypeScript. The agent orchestrates four complementary external tools: Apollo for international contacts, Pappers for French firmographic data (size, fundraising, directors), Unipile for LinkedIn signals, and Perplexity for open web research on unusual criteria. It reads the prompt, extracts explicit and implicit filters, dynamically selects the providers to query, merges and qualifies the results.

First attempt on OpenAI's Assistants API, quickly discarded: as soon as you need to chain tools based on intermediate results, it becomes a mess. We switched to Agents SDK the moment it launched. Cleaner orchestration, dynamic tool routing, native observability. The agent reasons about its own API calls instead of running a script.

Feature shipped in two weeks, natively integrated into Sinky's product. Users express their need in one sentence and get between 10 and 25 qualified prospects in about ten seconds, ready to go into a multichannel campaign. A new product differentiator in a market saturated with filter-based tools.