Boruna Enterprise Platform Overview

Boruna is a deterministic execution platform for enterprise AI workflows. It provides policy-gated, auditable workflow execution with built-in governance, replay, and compliance evidence generation.

Core Capabilities

Workflow Execution

• DAG-based workflow definitions with typed data flow between steps
• Topological execution ordering with dependency resolution
• Approval gates for human-in-the-loop review
• Retry policies for transient failures
• Budget enforcement per step and per workflow

Determinism Guarantees

• Same inputs + same workflow + same policy = identical outputs
• BTreeMap-based ordering throughout (no HashMap non-determinism)
• Capability-gated side effects — all IO is declared and controlled
• Record/replay support via EventLog

Policy Enforcement

• Capability allowlists: declare which side effects each step may use
• Budget limits: token and call count budgets per step
• Model allowlists: restrict which LLM models may be invoked
• Network allowlists: restrict outbound HTTP destinations

Audit Trail

• Hash-chained audit log: tamper-evident, append-only record of all decisions
• Evidence bundles: self-contained compliance artifacts per workflow run
• Bundle verification: cryptographic integrity checking of all artifacts

Architecture

workflow.json → Validator → Runner → Evidence Bundle
                  ↓           ↓
              Topological   Per-step:
              Sort          Compile .ax → VM → Output
                            Policy check
                            Audit log entry

Crate Map

• boruna-orchestrator — Workflow engine, audit system, evidence bundles
• boruna-compiler — Compiles .ax source to bytecode
• boruna-vm — Executes bytecode with capability gating
• boruna-bytecode — Bytecode format and Value types
• boruna-effect — LLM integration with budget tracking
• boruna-framework — App protocol (Elm architecture)
• boruna-tooling — Diagnostics, repair, trace-to-tests, templates
• boruna-pkg — Package system with integrity verification
• boruna-cli — The boruna command-line binary (workflow, evidence, lang, framework, trace2tests, …)
• boruna-mcp — MCP server binary exposing the toolchain to AI coding agents over stdio
• boruna-lsp — Language server for .ax files (diagnostics, completion, formatting)

Workflow Lifecycle

1. Define — Write workflow.json with steps, edges, policies
2. Validate — boruna workflow validate <dir> checks DAG structure
3. Run — boruna workflow run <dir> --policy <policy> executes steps
4. Record — --record flag generates evidence bundle
5. Verify — boruna evidence verify <dir> checks bundle integrity
6. Replay — There is no workflow-level replay command. For a single .ax program, boruna run <file> --record <log.json> saves the VM event log and boruna replay <file> <log.json> re-runs the program (.ax or .axbc) against the recorded capability results and reports whether the replayed event log matches.

Schema Versioning

Versioned formats and their version fields:

• Workflow definition: schema_version: 1 (required; unsupported versions are rejected)
• Policy: schema_version: 1
• Evidence bundle manifest: schema_version: 1 plus format_version: "1.1" (readers accept any 1.x, reject a different major)
• VM event log (EventLog): version: 2
• Audit log: no version field. A verifier detects the entry form per entry: 1.1 commitment-chain entries carry content_sha256, legacy 1.0 entries do not

Edit this page on GitHub