This is great, question though, since you support coding agents like Claude Code and Codex, they run MCP servers, plugins, web search/fetch and their own telemetry from the dev's machine. Do you give customers guidance or controls for that side, or is it out of scope for the gateway?
ericmags 50 minutes ago [-]
We do give our customers guidance to make sure they are thinking about this properly. This typically happens during on-boarding.
We also built out pre-defined controls for common tools so they can click a button and not think about it.
projectescape 42 minutes ago [-]
Food for thought, tool and MCP calls run locally, but every request carries the tool list and every response carries the model's tool calls, both of which pass through you. You could strip non-approved tools from the request and block calls to them in the response, based on an org allowlist. That gives you org-approved MCP/tool calls without touching the dev's machine.
ericmags 21 minutes ago [-]
Great call. We already block certain tool schemas and flag URLs in tool-call arguments, and since we enforce org allowlists for models on every request, extending that to tools/MCP is a natural next step.
skinfaxi 45 minutes ago [-]
How does ZDR work with requests that trip security measures in the model provider? Do the providers retain those requests for analysis or is that too guaranteed?
ericmags 23 minutes ago [-]
The infra providers call this "abuse monitoring" or "abuse detection" and we've worked with them to turn it off. We don't serve models like Fable that require it. We have one case remaining for OpenAI's GPT-5.x models in AWS. AWS's automated classifier can retain a flagged request for up to 30 days. Per AWS, that review is automated with no human access, the data stays in AWS GovCloud, and it's never shared with OpenAI. We are actively working to turn that off as well. Everything else is covered.
runamuck 2 hours ago [-]
I love this. Will this guarantee that proprietary/ sensitive/ CUI data won't leak to model training or other parties?
ericmags 2 hours ago [-]
Correct. We have contractual agreements to ensure no data that passes through our platform will be used for model training purposes. In addition, we have in our contracts that we (Consus) will also not use any data for training purposes. We couldn't if we wanted to since we do not store any prompt data.
jayonsoftware 42 minutes ago [-]
Does this work with HIPAA data ?
ericmags 14 minutes ago [-]
We've been focused on defense, so we haven't set up HIPAA (BAAs) yet. The underlying controls are essentially FedRAMP High, so it wouldn't be a big lift on our side. If you have a specific use case in mind, drop me a note at founders@consusindustries.co and we can talk more about it.
rahibali 2 hours ago [-]
[flagged]
ericmags 2 hours ago [-]
Glad you like the compliance suffix.
W.r.t observability: We don't use any third-party error tracker, APM, or queues. If any request fails, the client gets an error message that we write as opposed to the raw provider error. On our side, every log lives inside our boundary, and is written from a field allowlist (ids, model, token counts, status, latency). The JSON is purposely stripped away so any details in the request do not land in our logs.
MisterMunchkin 1 hours ago [-]
What is this AI slop even supposed to mean? Did the company write this?
Rendered at 17:55:13 GMT+0000 (Coordinated Universal Time) with Vercel.
We also built out pre-defined controls for common tools so they can click a button and not think about it.
W.r.t observability: We don't use any third-party error tracker, APM, or queues. If any request fails, the client gets an error message that we write as opposed to the raw provider error. On our side, every log lives inside our boundary, and is written from a field allowlist (ids, model, token counts, status, latency). The JSON is purposely stripped away so any details in the request do not land in our logs.