For technical teams
Under the hood.
A team harness, built on a real data model.
Built by people who build agents.
Operalta is an agent-first platform. The app, the API, the MCP server and the CLI share the same data and permissions, so your own agents work on the same ground.
~ $ claude
> Run the monthly re-score on the Hydrogen matrix
⏺ operalta · activate_goal_loop (“Re-score · Hydrogen”)
⎿ 38 entries · stops for approval before stage 3
Made for teams.
Most agent tools serve one person. Here the team works in shared conversations with a shared memory, and every agent acts with the rights of the person it works for.
Knowledge as a graph, memory as events.
Knowledge is linked as a graph, and changes are kept as events in time. That is what we are building on so the AI can reason about causes and look ahead, not just retrieve documents.
One data model.
Decisions, evidence, goals and work are connected records, not files. Each organization is isolated at the database level.
A workflow engine on the same data.
Any matrix can become a governed workflow: stages, allowed moves and approvals.
Agents that keep going.
Runs survive interruptions, sub-agents work in the background, and every run knows who or what started it.
Approvals that hold.
Actions you mark as risky wait for a person. If the target changes after the request, the approval no longer applies.
Every action counted once.
Writes to connected tools are logged as they happen. Anything uncertain goes to a person.
Learning with a test.
A new skill must beat the current one on real past cases before it ships. Lessons reach the team only with an approval.
Open by default.
MCP, CLI and API. A plugin for Claude Code and Codex. 14 model providers, several in Europe, or bring your own key.
Hundreds of improvements shipped every week.