Key Features
This page maps Wardian's major capabilities to their detailed guides.
Malleable Agent Environment
- Local-first agent habitat where prompts, classes, skills, workflows, queue evidence, and memory-ready context are inspectable artifacts rather than opaque app-only state
- Multiple lenses over shared Wardian state: desktop UI, CLI, Library, Workflows, Explorer, Queue, Graph, and future spatial views operate against the same canonical records
- Gentle slope from use to creation: run an agent, save a prompt, tune a class, deploy a skill, automate a workflow, then promote durable evidence into memory or project context
- Explicit scopes for global, class, agent, team/project, workspace, and workflow-run context so reusable capabilities can be shared without losing provenance
Related docs:
Live Agent Habitat
- High-density Grid and Dashboard views for live local agent sessions
- Left control rail for spawning, commands, workflows, Explorer, and settings
- Right roster with status lights, thought snippets, and agent working sets
- Queue tab for unread agent completions and workflow outcomes
- Roadmap direction: Sites, Cohorts, movable surfaces, Garden spatial organization, and Graph communication topology
Related docs:
Agent Classes and Reusable Library
- Class blueprints with instruction files and assigned skills
- Prompt and skill management from a filesystem-backed library
- Starred prompts for one-click execution from the Command panel
Related docs:
Broadcast and Prompt Injection
- Broadcast text commands to selected agents or all active agents
- Inject starred prompts from the library after flattening into terminal input
Related docs:
Agent-Facing Wardian CLI
- Give agents a textual control surface for inspecting live or persisted Wardian state
- Let agents spawn, clone, pause, resume, kill, wait on, and watch peers through the same backend control layer used by the desktop app
- Send prompts from inline text, standard input, or files; use
wardian askfor bounded peer handoffs with response evidence - List workflows, show workflow definitions, and start or stop workflow runs when the desktop app is available
- Inspect Wardian-managed agent worktrees, teams, and watchlists for automation-friendly coordination without treating the CLI as the primary human UI
Related docs:
Source Control Per Agent
- Git status, staging/unstaging, discard, commit, pull/push
- Inline diff viewing and commit history
- Optional worktree mode for branch isolation
Related docs:
Workflow Builder and Runtime
- Visual node-based builder with variable assistant
- Manual, scheduled, and listener-style trigger behaviors
- Scheduled run management and run-time role assignment
- Deterministic Rust workflow engine with pulse-driven candidate-node execution, branch/loop/wait control, agent execution modes, shared storage, and live telemetry
Related docs:
Queue and Completion Triage
- Captures agent completions when active terminal output settles back to Idle
- Records completed and failed workflow outcomes for later review
- Persists queue items under the Wardian home so unread work survives app restarts
- Supports unread badges, mark-read, clear-read, dismiss, and expandable long summaries
Related docs:
Runtime Shell and Session Policy Controls
- Default shell selection (
auto, discovered shell, or custom executable) - Per-installation regular-agent session persistence policy (
resumevsfresh) - Theme sync support and Gemini patch actions from settings
Related docs:
Provider-Aware Runtime
- Unified runtime model over Antigravity, Claude, Codex, and OpenCode (Gemini is unmaintained — use Antigravity for Google-model access)
- Provider-specific lifecycle handling for resume/session identity and skill discovery
Related docs: