Intent-Engine

中文 | English

Persistent memory for AI coding assistants.

AI Forgets. Every Time.

Without Intent-Engine:

Day 1: "Build authentication"
       AI works brilliantly...
       [session ends]

Day 2: "Continue auth"
       AI: "What authentication?"

With Intent-Engine:

Day 1: "Build authentication"
       AI works, saves progress...
       [session ends]

Day 2: "Continue auth"
       AI: "Resuming #42: JWT auth.
            Done: token generation.
            Next: refresh tokens."

One command restores everything: ie status

Visual Dashboard

See your entire task structure at a glance:

ie dashboard

Features:

• Task Navigator — Hierarchical tree view with search
• Task Detail — Full spec with markdown rendering (mermaid diagrams, code blocks)
• Decision Timeline — Chronological log of all decisions and notes
• Multi-project Support — Switch between projects via tabs

Not Just Memory — Infrastructure

What actually happens when things go wrong:

• Session ends → ✓ Persisted
• Tool crashes → ✓ Recoverable
• Week later → ✓ Full history
• Multiple agents → ✓ Isolated
• Complex project → ✓ Focus-driven

Why It Works

Minimal Footprint — ~200 tokens overhead, single binary, no daemons

Battle-Tested Stack — Rust + SQLite + FTS5, GB-scale in milliseconds, local-only

The Bigger Picture

The unsolved problem in AI agents: tasks that span days or weeks.

Intent-Engine provides the foundation:

Week-long refactoring:

├── Agent A (session: "api")    → focus: #12 REST endpoints
├── Agent B (session: "db")     → focus: #15 Schema migration
└── Agent C (session: "test")   → focus: #18 Integration tests
                                  depends_on: [#12, #15]

• Interruptions → Persistent memory
• Multi-agent → Session isolation
• Scheduling → Dependency graph (depends_on)
• Context explosion → Focus-driven retrieval

Result: Reliable multi-day, multi-agent workflows.

Get Started

Claude Code (recommended)

/plugin marketplace add wayfind/origin-task
/plugin install intent-engine

OpenCode

curl -fsSL https://raw.githubusercontent.com/wayfind/intent-engine/main/opencode-plugin/intent-engine.ts \
  -o ~/.config/opencode/plugin/intent-engine.ts

Manual Install

# Choose one
npm install -g @origintask/intent-engine
brew install wayfind/tap/intent-engine
cargo install intent-engine

# Or use the install script
curl -fsSL https://raw.githubusercontent.com/wayfind/intent-engine/main/scripts/install/ie-manager.sh | sh -s install

Core Commands

ie status                         # Restore context
ie dashboard                      # Open visual dashboard
echo '{"tasks":[...]}' | ie plan  # Create/update tasks
ie log decision "chose X"         # Record decisions
ie search "keyword"               # Search history

LLM-Powered Features (Optional)

Event-to-Task Synthesis - Automatically generate structured task summaries from event history:

# Configure LLM (one-time setup)
ie config set llm.endpoint "http://localhost:8080/v1/chat/completions"
ie config set llm.api_key "sk-your-key"
ie config set llm.model "gpt-3.5-turbo"  # Or local model

# Test connection
ie config test-llm

# Now when completing tasks, synthesis happens automatically for AI-owned tasks
ie task done 42  # Generates structured Goal/Approach/Decisions/Outcome summary

Cost Awareness:

• 1,500 tokens per synthesis ($0.003 with GPT-3.5-turbo)
• 20 tasks/day ≈ $22/year with GPT-3.5, or use local models (free)
• Synthesis only happens when LLM configured (graceful degradation)
• See LLM Use Cases for full details

How It Works

Session Start  →  ie status  →  Full context restored
                                       ↓
Working        →  ie plan    →  Tasks tracked
               →  ie log     →  Decisions recorded
                                       ↓
Interruption   →  Auto-persisted
                                       ↓
Next Session   →  ie status  →  Continue where you left off

Documentation

• Quick Start — Get running in 5 minutes
• Dashboard Guide — Visual interface walkthrough
• CLAUDE.md — AI integration guide
• Commands — Full reference

MIT OR Apache-2.0 · GitHub

Give your AI the memory it deserves.