Three months ago, I was convinced we were doing something wrong.
At Attri, we were deep into building AI-powered workflows, agents that could research prospects, summarize calls, draft follow-ups, close the loop on deals. The agents were sharp. The reasoning was solid. But every single workflow kept breaking at the same place.
Where does the agent actually put what it knows?
Not summarize. Not generate. Store. Reliably. Traceably. In a way that another agent, or a human, could audit three weeks later and understand exactly why it believed what it believed.
We pointed our agents at Salesforce. Then HubSpot. Then a handful of integrations that promised "AI-ready" everything. Same result every time. These tools were built for a human who logs in, clicks through tabs, types into fields, and hits save. Our agents don't do any of that. And no amount of API wrappers changes the fundamental architecture underneath.
The breaking point was a deal that fell apart because two agents had conflicting information about budget — and neither of them knew the other existed. One had read an email from two weeks ago. The other had read a call transcript from yesterday. The CRM just stored the last update and moved on. No conflict. No flag. No trace. Just a wrong number sitting in a field, looking like truth.
That's when we stopped trying to adapt existing tools and started drawing on a whiteboard.
What would a CRM look like if the agent was the primary user, not the human? No UI needed. No click-through flows. Just a clean, typed, structured data layer with one job: give agents reliable, verifiable, traceable memory.
We built it. We called it Open Tooling CRM. And last Friday, we launched it open-source.
The core idea is the evidence chain. Every claim an agent makes traces back through a structured path, raw artifacts like emails and transcripts, typed observations extracted from those artifacts, derived summaries that cite those observations, and explicit conflict records when observations disagree. No silent overwrites. No opaque AI black boxes. Every claim has receipts.
It's headless by design, no frontend ships with it. 29 REST endpoints, 27 MCP tools for Claude and any MCP-compatible client, SQLite-backed, fully self-hosted. Three commands to run locally.
But honestly, the technical spec isn't the point of this post.
The point is that we spent months assuming the infrastructure problem was ours to solve around. It wasn't. The infrastructure itself was wrong. And if you're building agents that touch any kind of business data, CRM, HR, finance, project management, you've probably felt this too. The agents are ready. The data layer underneath them isn't.
Open Tooling is our answer to that. CRM is the first module. There's a lot more coming.
If any of this resonates, the repo is open, Apache 2.0, and we'd genuinely love to hear what you're building.
GITHUB: github.com/Attri-Inc/open-tooling
PLUGIN MARKETPLACE: github.com/Attri-Inc/open-tooling-plugins
