Show HN: Nous – give GTM agents one context graph across your tools(github.com)
github.com
Show HN: Nous – give GTM agents one context graph across your tools
https://github.com/NousC/opennous
7 comments
This is crazy! Have you been talking to Claude?
I came to the realization that I needed a identity graph last week and started testing it one in Parallel AI.
We have 5 tools where customer touchpoints happen and we had to manually stitch them together for the full view.
Congratulations on the launch, ill check it out today.
I came to the realization that I needed a identity graph last week and started testing it one in Parallel AI.
We have 5 tools where customer touchpoints happen and we had to manually stitch them together for the full view.
Congratulations on the launch, ill check it out today.
Hi, we would love your feedback and what more you expect from your context layer. Please feel free to reach out at [email protected]
This came from an idea that had been knocking in my head for everybody as I open the browser. I tried different styles but hard to believe that there are close to 60 so hopefully you can refresh to few
It's open source under the AGPL license, so you can self-host it. Happy to discuss identity resolution, the epistemic model (observations to entities to claims), and answer questions.
It’s really impressive, on what basis you are computing the confidence score to the observations?
It's deterministic math: when two observations agree, our confidence goes up. If they conflict, confidence drops. Over time, a decay function lowers confidence as facts get older.
How are you solving identity resolution? What happens when two people share a name at the same company?
It's a waterfall sequence of hard identifiers such as email, LinkedIn URL, domain, and CRM ID and it only accepts exact matches and does not use fuzzy search. This approach works for most situations.
If there is still no match, the system creates a new entity instead of making an incorrect guess.
How do you handle conflicting facts from two sources when you are extracting the observations?
The most recent observation is used as the claim. If there are conflicting observations, each one lowers the confidence score, but the newest still takes priority. The only exception is that CRM fields follow their own priority order. The human-entered data comes first, followed by enriched data, then inferred data. For all other cases, the newest observation is used.
how is this different from just giving my agent access to my CRM?
CRMs are built for humans to update records. Nous is built for AI agents to read and write context. It stores observations, derives claims, tracks confidence and freshness, and gives agents one API instead of making them coordinate multiple tools.
so I would still need a CRM but agents would use the context graph?
Exactly, CRM is for your business records and the context graph is for aligning your agent!
Nous is a context layer beneath your agents that turns each touchpoint into a structured observation linked to the right person and company. It derives claims, which are beliefs about a fact with their confidence and freshness. The graph runs on your own data, improves with each agent outcome, allowing one agent to read the whole account in a single call instead of using six tools.