Your Agent — Quick Reference

Your agent is a folder (agent.yaml + instructions/ + workflows/ + knowledge/). The Knowledge Engine (@soleri/core) provides all the persistent infrastructure below. Edit your agent’s files — the engine handles the rest.

This page is your cheat sheet. For detailed explanations, see the linked deep dives.

The Vault

Your agent’s long-term knowledge store. SQLite database with full-text search, branching, sharing, and multi-tier connections. Details →

Entry structure:

Field	Values
Type	pattern, anti-pattern, rule, playbook, workflow, principle, reference
Domain	frontend, backend, security, or your custom domains
Severity	critical (must follow), warning (should follow), suggestion (nice to have)
Tags	Free-form labels for discovery

Common operations:

"Search for authentication patterns"
"Capture this pattern: always use error boundaries at route level"
"Show me vault stats"

Advanced features: vault branching (experiment without affecting main vault), Obsidian sync, knowledge pack import/export, team review workflows, multi-tier vault connections. 66 ops →

The Brain

Tracks which patterns actually work. Learns from usage, strengthens useful patterns, decays unused ones. Details →

What it does:

• Ranks search results by proven usefulness, not just keyword match
• Surfaces recommendations when you create plans
• Extracts patterns automatically from completed work sessions
• Runs a full lifecycle: extract → promote → archive

Common operations:

"What does the brain recommend for this task?"
"Show me pattern strengths"
"Rebuild brain intelligence"

Memory & Cross-Project

Knowledge persists across sessions in local files. Link projects to share knowledge across codebases. Details →

Common operations:

"Link this project to ../api-server as related"
"Search across all projects for deployment patterns"
"Promote this pattern to global"

Planning

Multi-step task planning with grading, verification, evidence, and drift detection. Details →

The lifecycle: create → grade → approve → split → execute → reconcile → complete

Common operations:

"Create a plan for migrating the auth system"
"Grade this plan"
"Submit evidence for task completion"
"Reconcile the plan against what happened"

32 ops →

Playbooks

Multi-step procedures with validation criteria at each step. Details →

Common operations:

"List available playbooks"
"Run the API endpoint review playbook on this code"
"Create a playbook called 'Database Migration' with steps: ..."
"Find a playbook that matches this context"

8 ops →

Orchestration

Plan → Execute → Complete lifecycle for complex tasks. Brain recommendations feed into plans, completed plans extract knowledge back to the vault. Details →

The compound loop:

vault knowledge → brain recommendations → plans → work → knowledge extraction → vault

Governance

Controls how knowledge enters the vault — quotas, proposal gates, duplicate detection. Details →

Presets: strict (all require approval), moderate (auto-approve suggestions), permissive (auto-approve all)

Curator

Automated vault quality management — deduplication, contradiction detection, health audits, tag normalization, LLM enrichment. Details →

Common operations:

"Run a health audit"
"Detect duplicates in the vault"
"Check for contradictions"
"Groom all vault entries"

13 ops →

Chat

Session management, authentication, voice, browser isolation, and message queue for chat transports (Telegram, web). 41 ops →

Subsystems: sessions, auth, MCP bridge, task cancellation, self-update, file handling, notifications, voice (Whisper + TTS), message queue, per-chat browser isolation.

Agency

Proactive mode — your agent watches file changes and surfaces relevant vault patterns without being asked. 8 ops →

"Enable agency mode"
"Scan this file for warnings"
"Show pending warnings"

Context

Entity extraction, knowledge retrieval, and context analysis. Analyzes prompts to extract files, functions, domains, and technologies. 3 ops →

Transports

Four ways to connect: stdio (your AI editor), HTTP/SSE (web), WebSocket (real-time), LSP (editors). Guide →

All 13 Facades

Facade	Ops	Primary purpose
Vault	66	Knowledge storage, search, branching, sharing
Admin	56	Health, telemetry, plugins, packs, accounts
Chat	41	Chat transport integration
Plan	32	Planning, grading, verification
Orchestrate	26	Lifecycle, projects, playbooks
Brain	30	Learning, strength, recommendations, radar
Memory	15	Cross-session, cross-project
Curator	13	Vault quality management
Control	13	Identity, governance
Archive	12	Vault archival, lifecycle, maintenance
Operator	10	Profile learning, signals, adaptation
Loop	9	Iterative validation
Links	9	Entry linking, graph traversal
Agency	15	Proactive file watching, intelligence
Sync	8	Git, Obsidian, pack sync
Intake	7	Content ingestion (URLs, text, books)
Tier	7	Multi-vault connections
Branching	5	Vault branching, merge
Review	5	Knowledge review workflow
Context	3	Entity extraction, analysis

Total: 350+ operations plus 5 per domain.

Day-to-Day Tips

1. Follow the workflow — Search → Plan → Work → Capture → Complete
2. Capture as you go — the moment you learn something, capture it
3. Search before building — 5 seconds of search can save hours of rework
4. Use domains — keep knowledge organized so searches stay relevant
5. Review brain recommendations — they reflect what actually works in your project

Your data stays on your machine — Security & Privacy. For term definitions, see the Glossary. If something isn’t working, check Troubleshooting.