All case studies
    European Space AgencyCase study · Space operations

    When a spacecraft misbehaves, answers can’t take hours.

    We built OCAI Next Generation for the European Space Agency: a multi-agent assistant that helps ESOC operators resolve anomalies faster — with every answer traced back to the ESA documentation and telemetry behind it.

    See how it works
    OCAI NG · mission-operations workspace
    OCAI NG answer view with cited ESA sources
    Client
    European Space Agency (ESA)
    ESOC, Darmstadt · 2024–2025
    15
    institutional sources in one assistant
    20
    missions using it at ESOC
    Hours → min
    telemetry-to-procedure correlation

    Your challenge

    The search worked. The analysis was still manual.

    During a spacecraft anomaly, operators cross-referenced live and historical telemetry, dense Flight Operation Procedures and anomaly databases by hand.

    ESA already ran OCAI 2.0, with semantic search across 15 institutional sources. But the sheer volume of results still forced lengthy manual analysis — and every extra minute pushed up Mean Time to Resolution.

    The brief: add AI on top of what already worked, not start again.

    Live and historical telemetry

    Thousands of parameters to compare against past behaviour.

    Flight Operation Procedures

    Dense, safety-critical documents to find and interpret.

    Anomaly databases

    Has this happened before — and what fixed it?

    Our solution

    Search returned documents. Agents return answers.

    OCAI NG is a multi-agent assistant layered onto the existing mission-operations workspace. Operators ask in plain language and get one synthesised answer — not a pile of results to read through.

    Before · OCAI 2.0
    • One query, results grouped source by source
    • Keyword search downstream of the search server
    • Operators connect telemetry, procedures and past anomalies themselves
    • Correlation takes hours
    After · OCAI NG
    • A coordinator routes each question to specialised agents
    • Telemetry, procedures, anomaly records, logs, JIRA and ARES synthesised into one answer
    • Every statement cites the ESA source behind it, so operators verify before they act
    • Correlation takes minutes

    How it works

    One question. Specialist agents. One answer you can verify.

    OperatorAsks in plain language during an anomaly
    Coordinator agentPlans the analysis and routes the question
    Telemetry agentLive and historical parameters
    Procedures agentFlight Operation Procedures
    Anomaly agentAnomaly databases and ARES
    Logs agentOperations logs and JIRA
    Grounded answerOne synthesised response, every claim linked to the ESA documentation and telemetry it came from

    Simplified view. The full MVP architecture is shown in the project gallery below.

    Built for mission operations

    Where a confident wrong answer is worse than no answer.

    Grounded by design

    Enterprise agentic RAG with strict institutional guardrails. Every response cites its sources, which reduces hallucination risk and keeps operators in control.

    Nothing leaves ESA

    Runs fully on-premise on open-source models. Mission data stays inside ESA’s own infrastructure — no external AI provider involved.

    Layered, not replaced

    Built on top of OCAI 2.0 and the platform that already indexed ESA’s institutional sources. Operators kept the tools they trust.

    TechnologyPythonKubernetesOpen-source LLMsMulti-agent systemsLangbox, our own platform

    Inside the project

    From grouped results to grounded answers.

    The starting point: OCAI 2.0 — one query, results grouped source by source, with no synthesis.
    01The starting point. OCAI 2.0: one query, results grouped source by source, no synthesis.
    Before OCAI-NG: the IOA platform indexed the institutional sources, but everything downstream of the search server was keyword search.
    02Before OCAI NG. Sources were indexed, but everything after the search server was keyword search.
    OCAI NG: a multi-agent assistant layered onto the mission-operations workspace — grounded answers with verifiable citations.
    03OCAI NG in the workspace. Grounded answers with citations operators can check.
    The OCAI-NG MVP component architecture — what already existed, and what we built new for OCAI-NG.
    04The MVP architecture. What already existed, and what we built new.

    The results

    In production across ESOC — not stuck in a pilot.

    Hours → minutes

    Telemetry-to-procedure correlation, the slowest step in anomaly analysis.

    20 missions

    In use at the European Space Operations Centre.

    Every answer cited

    Reduced hallucination risk: responses point to ESA documentation and telemetry.

    Measured

    Adoption and time-to-answer tracked mission by mission.

    Stop searching.
    Start resolving.

    Your experts lose hours to documentation, live data and past cases spread across systems? Let’s talk about what grounded, on-premise agents can do for your operations.

    AI Automation and Agents