No vendor lock-in.
Your data stays within your perimeter. No HOLLI license.
The pillars
- DeploymentOn-prem or private cloudYour data stays at your company
- AgentHermes or anotherAuditable and interchangeable: nothing locks you into a single tool
- LLMOpen weights modelsLocal or private cloud (ZDR)
- OrchestrationMCP + agent specializationEach agent is tailored to your workflows, not a generic assistant
MCP (Model Context Protocol): the open standard that lets agents talk to your tools, instead of a proprietary integration per tool.
Concretely, this stack relies on well-known open source tooling: Ollama and LM Studio for running models locally, Mistral, Qwen, Minimax or Kimi as open weights models depending on your constraints, Hermes Agent, OpenCode or Openclaw as the agent layer. The exact choice is made at the first agent: this page says nothing is proprietary, not which specific tool you will get.
Depending on what the first agent reveals, we also configure private and/or sovereign cloud: providers like OVH or Scaleway, with a Zero Data Retention (ZDR) option: the provider keeps no trace of what passes through its servers.
Interoperability
FTrack · ShotGrid · Kitsu · Asana · Slack · Gmail · Teams · Excel…and what you already use.
We integrate into your control and decision system, we don't rebuild it.
What gets pinned down at the first agent
This page sets out the principles. The details (exact configuration, model choice, connectors to build, guardrails and agent access rights) are priced and written at the first agent (step 3), on your real IT system, not a theoretical one.
What we can already say: it runs on CPU, with a speed gain on GPU; sizing depends on your volumes. The guardrails (what an agent can do, what it can't do, and how that is controlled) are defined and documented at the first agent, never left at defaults. What leaves your perimeter (updates, telemetry and licenses) is listed at the first agent, before any deployment: nothing gets added afterwards without your agreement.