Take back control of your company’s AI.

Your teams already use AI, in accounts nobody can audit. darkplum puts it inside your own cloud account or data center, where every prompt follows your policy and the knowledge stays yours.

  • Data residency you can prove: self-hosted and Bedrock models never leave your infrastructure
  • Knowledge that stays yours: skills, tools and history in your own database
  • One policy for every team, every key, every request
Admin > Models > Policy: where requests may run, with the organisation held to its own cloud account and vendor APIs refused

Governance, visibility and control, before the first prompt is sent.

Governance

Policy decides where a prompt may go. The organisation sets how far, users and keys can only be held tighter, and vendor APIs stay off until an administrator turns them on.

Visibility

Every model is labelled by where it runs. Every request is recorded with the class that answered, refusals included. Administrators see who, when and which model, never the text.

Control

Allowlists per user and per key. Tools behind an administrator’s host allowlist, with credentials in your cluster. A library that exports as plain files the day you want to leave.

See how it works.

Every screen is the product as it runs today, captured on a local cluster.

Admin > Models: the catalogue grouped by backend, each model labelled self-hosted or cloud-managed, with a count of models that send prompts outside the account

Today

Model names say nothing about whose servers answer them, or how many prompts are leaving.

With darkplum

Every model labelled by where it runs. Every request recorded with what answered.

Admin > API keys: keys scoped to a user, one pinned to self-hosted models only

Today

A developer’s key can reach more than the developer is allowed to.

With darkplum

Keys inherit their owner’s limits and never exceed them.

Admin > Routing: a routed alias with three intent routes, each pointing at a self-hosted model, and the in-cluster semantic classifier

Today

Every prompt goes to the most expensive model because nobody picks.

With darkplum

One name, routed by intent to the right model, with the classifier in your cluster.

A tool in the library: its MCP server, the verified tool catalogue and a stored credential

Today

Connecting an internal tool means handing a vendor the credential that reaches it.

With darkplum

Tools behind an administrator’s allowlist, credentials in your cluster.

The Library page: skills, context and tools, some private and some shared with the organisation

Today

Prompts, instructions and reference notes live in one vendor’s account, one person at a time.

With darkplum

A library in your own database, shared across the organisation.

darkplum chat: a postmortem drafted with a skill and a wiki tool from the library, on a self-hosted model

Today

Shadow AI exists because the sanctioned option was worse.

With darkplum

A chat your people will use, with history only its owner can read.

Install it yourself.

One engineer, one sitting. You need the AWS CLI, Terraform 1.9 or later, kubectl, jq, Docker and Go. Set the region, profile and GPU node count in the variables file before the last command.

About twenty minutes and roughly $305 a month at list price without a GPU node; 35 to 50 minutes and about $1,035 with one.

Read the full install guide

$ cd darkplum
$ make cli && ./bin/darkplum doctor check --cloud aws --profile <your-profile>
$ cp terraform/aws/terraform.tfvars.example terraform/aws/terraform.tfvars
$ make deploy

Or on your laptop, with Docker and kind and no cloud account:

$ cd darkplum
$ make dev-up