Skip to content

AIX overview

AIX is Kesita’s legal workspace assistant. It combines conversation, private matter files, shared legal research, and editable documents while keeping the lawyer responsible for the final work product.

flowchart LR
    Lawyer[Lawyer] --> Workspace[AIX workspace<br/>chat, sources and editor]

    subgraph Product[Product domain]
        Workspace --> API[Product API]
        API --> ProductDB[(Product PostgreSQL)]
        API --> ProductBlob[(Private product blobs)]
        ProductDB -->|committed command| Worker[Product worker<br/>AIX capabilities + DBOS]
        Worker --> ProductDB
        Worker --> ProductBlob
        Worker --> Parser[Isolated document parser]
    end

    subgraph Legal[Legal domain]
        Worker -->|workload JWT| Core[Legal Core API]
        Core --> LegalDB[(Legal PostgreSQL + pgvector)]
        Core --> Corpus[(Corpus originals)]
        CorpusWorker[Corpus worker] --> LegalDB
        CorpusWorker --> Corpus
    end

    Worker --> Models[Model, embedding<br/>and OCR providers]
    CorpusWorker --> Models

The product and legal domains remain separate. AIX reaches shared legal material only through Legal Core. Private customer files remain in the product domain and never become part of the shared corpus implicitly.

flowchart LR
    Objective[Lawyer objective] --> Context[Authorized context]
    Context --> Plan[Bounded plan]
    Plan --> Research[Legal and private research]
    Research --> Draft[Draft or document proposal]
    Draft --> Verify[Evidence assessment]
    Verify --> Review[Lawyer review and acceptance]

AIX owns planning, tool selection, evidence assembly, drafting, and assessment. Deterministic application code owns authentication, authorization, budgets, task limits, revision checks, and publication.

AIX can read authorized data and create internal drafts, evidence, document proposals, and exports. It does not send email, file with courts, sign documents, make payments, or publish to third parties.

  • The Product API accepts commands and serves persisted workspace state.
  • Product PostgreSQL is authoritative for organizations, conversations, tasks, permissions, document revisions, usage, and ordered events.
  • DBOS runs inside the product worker and owns durable execution, retries, waits, and recovery—not chat history.
  • Legal Core owns legal retrieval and source delivery.
  • The corpus worker owns legal import, immutable source versions, embeddings, and atomic index publication.
  • Object storage owns file bytes; PostgreSQL owns their metadata, authorization, and lifecycle.
  • Model output is untrusted until schemas, permissions, evidence, and revision rules pass.

See Execution and data flow for the task lifecycle and Security and operations for enforcement and recovery.