Agentic

Agentic harness for LLMs in Swift. Provider-agnostic, and extensible ecosystem. Opt-ins to a ready-made CLI, tools, and skills. Or build your own.

Note: under development, unstable APIs

For a summary overview, see AI summary.

Index

Agentic ecosystem

• Agentic — Core provider-agnostic Agentic contracts.
• AgenticModels — Model, routing, request/response, and model-facing abstractions.
• AgenticTools — Tool definitions, discovery, exposure, and core tool infrastructure.
• AgenticSkills — Skills, skill metadata, loading, and skill infrastructure.
• AgenticWorkspace — Scoped workspace authority, roots, grants, and capabilities.
• AgenticIO — Agent-facing IO and source/context operations.
• AgenticExecution — Execution policy, preflight, approvals, prepared intents, and tool-plan execution.
• AgenticRuntime — Runtime lifecycle, sessions, coordination, suspension, recovery, and persistent execution.
• AgenticUsage — Usage, token, cost, and related accounting infrastructure.
• AgenticMedia — Media and multimodal Agentic capabilities.
• AgenticInterfaces — Human-facing interaction, host-console UI, approvals, inspection, and run control.
• AgenticAdapters — Model-provider adapters.
• AgenticDomains — Domain-specific Agentic integrations and tool sets.
• AgenticCLI — Ready-made command-line host for Agentic.

Significant supporting libraries

• Path — Paths, scoped paths, sandboxing, access rules, and scanning.
• IO — Filesystem and resource IO primitives.
• Readers — Typed source and resource reading primitives.
• Writers — Safe mutation planning, application, diffs, constraints, records, and rollback.
• FileTypes — File and resource type classification used across workspace and multimodal handling.
• Position — Canonical source positions, ranges, and line-aware coordinates.
• Selection — Source selections and ranges shared between search, reading, and context materialization.
• Concatenation — Source composition, selection, provenance, caching, and context materialization.
• Search — Deterministic source search, ranking, retrieval frontiers, and structural verification.
• Parsing — General parsing, lexing, grammar, and structural verification primitives.
• Parsers — Reusable parser implementations and higher-level parsing components.
• Tokens — Tokenization and token-aware source analysis primitives.
• Matching — Structured candidate matching and weighted match fields.
• Ranking — Deterministic scoring, ordering, thresholding, and result selection.
• Fuzzy — Fuzzy identifier and text matching used by discovery and retrieval.
• Primitives — Shared primitives including JSONValue.
• Schema — Schema types and @JSONSchema synthesis.
• Difference — Structured differences and diff representation.
• Arguments — Declarative command specifications, typed arguments, routing, parsing, and help rendering.
• TestFlows — Executable flow-based testing, assertions, diagnostics, snapshots, filtering, and test-run reporting.
• Guidelines — Structured guidelines and rationale metadata used to relate plans and actions to governing guidance.
• Milieu — Environment and runtime configuration primitives.
• Terminal — Reusable terminal interaction and TUI primitives.
• ANSI — ANSI styling and terminal representation.
• AWSConnector — Lightweight Swift AWS connectivity used by the Bedrock adapter.