Engagement
Three ways to engage.
From an independent evaluation to a bounded pilot to a multi-year build, each step deepens the commitment and the access to Plexus™ and the team behind it. Whichever one you choose, you decide how much autonomy your agents run with.
§ 01
Evaluation
Get an independent read.
About four to six weeks. We assess where audited AI agents fit your operations and map your governance posture against agentic-AI requirements. You get an independent read, and no platform commitment.
About 4 to 6 weeks
§ 02
Pilot
Prove it on one real workload.
About twelve to sixteen weeks. We stand Plexus up against one bounded workflow in your environment, with success criteria agreed up front. We install it on-prem, connect it to the agent tools your team already runs, such as Claude Code and OpenAI Codex, and enable your people, from GRC and security operations to the AI team, IT, and finance, to read the dashboard and act on it. Both sides have a clear exit.
About 12 to 16 weeks
§ 03
Design Partner
Build Plexus with us.
Eighteen to thirty-six months. Plexus is in production on your real workflows, with a dedicated AI engineer through enablement. Everything in the pilot, plus our harness, which carries your audited agents onto real applications and desktops, and your own on-prem pipeline to train or augment a local model on your business's work. You shape the roadmap, and you are the operating reference.
18 to 36 months
In every one, you choose the autonomy tier. Your agents run exactly as supervised or as independent as you decide, from actions approved before they run to fully air-gapped.
See the autonomy tiers →Design Partner exclusive
Your agents get smarter on your work. Privately.
As a Design Partner, your agents do not just run on Plexus. They improve on your work, without any of it leaving your environment.
Fully on-prem, fully private
The pipeline runs entirely in your environment. Your data, your agent trajectories, and your proprietary workflows never leave your infrastructure or touch a shared model. The model trained on your work is yours alone, never used to serve another customer.
Learns from real work
Plexus already records agent episodes. The pipeline curates and outcome-scores the best of them, trains or augments a local model on your own trajectories, and deploys the result back to your agents.
Better and cheaper on your tasks
A small model specialized on your workloads can match a large general one on the work you actually do, and run faster and cheaper doing it.
How it works
- 1
Capture
Plexus records agent episodes as they happen: the traces, the outcomes, the artifacts.
- 2
Curate & score
The pipeline selects the strongest episodes and scores them by outcome.
- 3
Train the adapter
It fine-tunes or augments a local model on your own trajectories, on your hardware.
- 4
Deploy
The specialized model goes back to your agents, better on the work you actually do.
Built on our own distillation workbench and a local-GPU training stack, deployable on your hardware.
Find the right way in.
Tell us how you run agents today and what you need to prove, and we will point you at the step that fits.