How Photon Built Live Lead Research Directly Into iMessage With TinyFish
At a glance
- Company: Photon, API infrastructure for bringing AI agents into the messaging surfaces people already use, including iMessage, WhatsApp, Telegram, and more
- Use case: Live research on a lead, surfaced inside the messaging thread before a rep replies
- Why: Accurate person resolution, sources attached to every claim, and fast enough to run before a rep starts typing
- TinyFish products: TinyFish Search, Fetch, and Agent power the research behind every outreach thread
- The split: Photon owns messaging, thread state, agent replies, and the handoff to a human. TinyFish resolves who is on the other end of the thread
Company overview
Photon is API infrastructure for agent messaging. Developers use it to put AI agents into iMessage, WhatsApp, Telegram, and similar channels, rather than asking people to download another app. The core is open source, and the documentation is here.
On top of the Photon iMessage API, the team built an outreach flow for sales teams. A rep texts a lead from a real business line, an AI agent handles the back and forth around the clock, a human takes over in one tap, and the whole team sees every thread in a dashboard.
TinyFish sits in front of the first message. Before a thread opens, it returns role, company, and social links for the lead, which is enough to personalize the opener and give the rep a tap-to-send first message. When a rep wants more, one tap reveals a fuller brief with sources attached.
The problem
The reason iMessage matters is conversion. SMS blasts get ignored, while an iMessage thread reads as a real person, with read receipts, replies, and genuine two-way conversation. Photon reports 60%+ better response than traditional SMS.
But texting someone is more personal than emailing them, which is exactly why it fails when you get it wrong. Photon's question was how to improve both response and retention on outreach, especially when an AI agent is the one sending the first message.
A form fill or an ad lead is worth reaching while intent is hot, which means minutes rather than hours. A rep opening a thread often has a name, maybe a company, and nothing else. Sending a personal message to someone you know nothing about is worse than sending nothing. Looking them up manually takes long enough that the moment passes.
The harder half of the problem is accuracy.
"Research scraping tools often collate profiles based on name, so getting an accurate read on the lead you are contacting can be difficult."
— Daniel Tian, Co-founder @ Photon
Names aren't unique. A tool that matches on name alone will confidently return the wrong profile, and the rep then opens with the wrong job title at the wrong company. That failure is worse than returning nothing, because a rep will use what they're given. In a channel that works because it feels personal, being confidently wrong costs more than being empty-handed.
Photon needed research that arrives before the rep starts typing, is sourced well enough to be trusted, and is careful enough to distinguish between people who share a name.
Why Photon built on TinyFish
Photon's core is open-source iMessage infrastructure. Deep web research sits well outside that, and doing it in-house would have meant owning crawling, extraction, and coverage across LinkedIn, X, GitHub, and the open web. None of that is what Photon sells. This was a new capability, not a migration from another research tool.
"Having TinyFish handle it let us ship a personalization layer on top of the messaging layer that we otherwise wouldn't have."
— Daniel Tian, Co-founder @ Photon
What the product asks of a web layer shows up in the output: role and company resolved from a name and a handle, social profiles found across LinkedIn, X, and GitHub, every claim traceable to where it came from, and enough coverage to notice when two people share a name.
TinyFish Search finds current sources. Fetch turns those pages into clean, usable content. Web Agent handles the deeper lookup when a rep taps for the full brief.
Photon keeps the conversation. TinyFish finds out who the person is.
Where TinyFish fits
A rep gets research at two points.
The first runs before the thread opens, so by the time the rep is looking at the conversation, the role, company, and social links are already there, along with a tap-to-send opener.
The second is one tap away, for when a rep wants more before replying. It returns a fuller brief: current role, public profiles, a source for each claim, and notes on anyone else who shares the name.
Both stay inside the messaging app. The research is stored against the lead rather than the device, so switching phones or passing a conversation to a colleague brings the brief along with the thread.
| Need | Photon | TinyFish |
|---|---|---|
| Who is this lead? | Shows the result in the thread | Search and Fetch resolve role, company, and social links |
| Make the first message personal | Drafts the opener in the lead’s own context | Returns current, source-backed details |
| Go deeper | One tap in the dashboard | Agent builds the fuller brief with sources attached |
| Keep the conversation going | Agent replies, human handoff, thread state | Research stays attached to the lead |
The outcome is a personal iMessage in seconds, referencing the lead's actual company and background, because the research underneath is fast and correct.
Key outcomes
Photon rates the difference 4 out of 5.
Reps open with the lead's actual company and background rather than a name in a template, and it takes seconds.
Every claim in the brief carries a source, so a rep can check it before using it.
What’s next
Photon wants to take the same research pattern into more verticals, starting with real estate, and into WhatsApp, where TinyFish already has a strong presence in Southeast Asia.
Building an agent that needs to know who it's talking to?
AI disclosure
Content on this website may be created or refined with the assistance of AI tools and is subject to human editorial review.
