Skip to main content
This page provides definitions for codebase-specific terms, jargon, and domain concepts within the Palyra ecosystem. It serves as a technical reference for engineers to map conceptual requirements to specific implementation details and data structures.

Core Concepts & Actors

Gateway Runtime (palyrad)

The central daemon process that coordinates all agent activities, tool executions, and storage operations. It acts as the hub between frontend applications (Web Console, CLI) and backend execution environments.

Agent Run Loop

The iterative process where the system alternates between LLM provider turns and tool execution batches.

The Tape

An append-only journal specific to an agent run. It records every event (user messages, tool calls, tool results, and model responses) to ensure deterministic replay and auditability.

Natural Language to Code Entity Mapping (Run Lifecycle)

This diagram illustrates how a natural language request from a user flows into specific code entities and state transitions within the daemon. Sources: crates/palyra-daemon/src/application/run_stream/orchestration.rs#1-15, crates/palyra-daemon/src/gateway.rs#72-85, crates/palyra-daemon/src/orchestrator.rs#136-137

Tooling & Execution

Sandbox Tiers

Palyra uses a tiered security model for executing code and processes to balance performance with isolation.

Workspace Scoping

A security constraint that limits tool access (filesystem, search) to specific directory roots, preventing path traversal attacks.

Tool Approval Posture

Determines if a tool requires explicit operator consent before execution.

Memory & Retrieval

Hybrid Retrieval

A search strategy combining lexical (FTS5) and semantic (Vector) search to find relevant context for the agent.

Instruction Compiler

A deterministic component that assembles the “System Prompt” for the LLM by layering runtime facts, tool schemas, and trust summaries.

Memory Scopes

Memory is segmented to prevent cross-contamination between users or sessions.

Security & Privacy

Redaction Pipeline

A multi-stage system that scrubs sensitive information (API keys, passwords, PII) before it is persisted to the journal or sent to the frontend.

Hash-Chaining

A tamper-evident mechanism where each journal event contains a SHA-256 hash of itself and the previous event’s hash.

Security Entity Mapping (Tool Execution)

This diagram maps the high-level concept of “Secure Tool Execution” to the specific code modules responsible for enforcement. Sources: crates/palyra-daemon/src/sandbox_runner.rs#1-15, crates/palyra-daemon/src/application/approvals.rs#84-85, crates/palyra-daemon/src/sandbox_runner.rs#121-122, crates/palyra-daemon/src/journal.rs#40-40

Technical Terms Reference

TermDefinitionCode Pointer
ACPAgent Control Protocol; bridge for CLI/TUI interaction.crates/palyra-cli/src/acp_bridge.rs#27-27
A2UIAgent-to-UI; the protocol for rendering rich UI components in the console.crates/palyra-daemon/src/gateway.rs#29-31
CronJobA scheduled agent run defined by a Recurrence rule.crates/palyra-daemon/src/journal.rs#91-92
FlowA multi-step agent procedure or lineage.crates/palyra-daemon/src/journal.rs#57-59
JournalStoreThe SQLite persistence layer for all daemon state.crates/palyra-daemon/src/journal.rs#7-9
ModelProviderAn adapter for external LLMs (OpenAI, Anthropic, etc.).crates/palyra-daemon/src/model_provider.rs#103-106
RecallThe process of retrieving relevant memories for a prompt.crates/palyra-daemon/src/application/recall.rs#101-101
VaultSecure storage for secrets (API keys) with envelope encryption.crates/palyra-daemon/src/gateway.rs#47-47
Sources: crates/palyra-daemon/src/gateway.rs#1-114, crates/palyra-daemon/src/journal.rs#1-165, crates/palyra-daemon/src/sandbox_runner.rs#1-153