Skip to content

Roadmap

Jump to bottom
jex edited this page Mar 10, 2026 · 1 revision

AISH Roadmap

Vision: Transform AISH from an intrusive AI shell into an intelligent sidecar assistant that gives users absolute control over their command line.

🎯 Strategic Goals (Q2 2026)

Core Principles

  1. User Control First: AI provides suggestions, never blocks the main workflow
  2. Async by Default: All AI analysis happens in the background
  3. Smart Intervention: Reduce false positives by 95% through intelligent filtering
  4. Progressive Enhancement: Maintain backward compatibility while evolving architecture

Success Metrics

  • ✅ Command execution returns to prompt in < 50ms
  • ✅ Zero blocking AI calls in main execution path
  • ✅ False positive rate < 5%
  • ✅ User satisfaction > 4.5/5

📅 Release Timeline

Week 1-2   │ v0.1.0 → v0.2.0  │ Async AI Sidecar Architecture
Week 3-4   │ v0.2.0 → v0.3.0  │ Smart Analysis & User Controls
Week 5-6   │ v0.3.0 → v0.4.0  │ Plan Mode & Enhanced Tools
Week 7-8   │ v0.4.0 → v0.5.0  │ Multi-Agent System
Week 9-10  │ v0.5.0 → v0.6.0  │ Rich UI & Task Management
Week 11-12 │ v0.6.0 → v0.7.0  │ Agent SDK & MCP Protocol

🚀 Phase 1: Architecture Overhaul (Week 1-4)

v0.1.1 (Week 1) - Critical Fixes

Type: Patch Release

Fixes

  • 🐛 Implement skill tool invocation (skill.py TODO)
  • 🐛 Fix tool call history not added to memory (shell.py:889)
  • 🐛 Disable auto-trigger of handle_error_detect() (temporary mitigation)

Impact: Reduces immediate user pain points

v0.2.0 (Week 2) - Async AI Sidecar 🔥

Type: Minor Release (Breaking Changes)

New Architecture

┌─────────────────────────────────────────────────────────┐
│                    Main Shell Process                    │
│  ┌──────────────┐                                       │
│  │ User Input   │ → Execute → Display → Prompt         │
│  └──────────────┘      ↓                                │
│                   Enqueue Event (non-blocking)          │
└─────────────────────────┼───────────────────────────────┘
                          ↓
┌─────────────────────────────────────────────────────────┐
│                  Background AI Sidecar                   │
│  ┌──────────────┐   ┌──────────────┐   ┌────────────┐ │
│  │ Event Queue  │ → │ Smart Filter │ → │ LLM Worker │ │
│  └──────────────┘   └──────────────┘   └────────────┘ │
│                          ↓                               │
│                   ┌──────────────┐                      │
│                   │ Notification │ → [AI:2] indicator  │
│                   └──────────────┘                      │
└─────────────────────────────────────────────────────────┘

New Modules

  • src/aish/sidecar/event_queue.py - Non-blocking command event queue
  • src/aish/sidecar/analyzer.py - Smart filtering logic
  • src/aish/sidecar/worker.py - Background LLM analysis worker
  • src/aish/sidecar/notification.py - Non-intrusive notification system
  • src/aish/sidecar/storage.py - Analysis result persistence

Key Features

  1. Event Queue: Commands enqueue completion events without blocking
  2. Smart Analyzer: Filters out false positives (grep, diff, ssh, Ctrl-C, etc.)
  3. Background Worker: Async LLM analysis in separate task
  4. Notification Center: Lightweight status indicator in prompt ([AI:2])

User Commands

  • :ai or :ai last - View latest suggestion
  • :ai list - List all pending suggestions
  • :ai show <n> - View specific suggestion
  • :ai apply <n> - Apply suggestion (enters security approval flow)

Breaking Changes

  • ❌ Removed auto-trigger of handle_error_detect()
  • ❌ Removed auto-trigger of handle_command_error()
  • ❌ Deprecated ShellState.CORRECT_PENDING

Migration Guide

  • Old behavior: AI automatically interrupts on command failure
  • New behavior: AI analyzes in background, user explicitly views suggestions
  • Compatibility: handle_command_error() API preserved for manual invocation

Configuration

# ~/.config/aish/config.yaml
sidecar:
  enabled: true
  max_queue_size: 100
  worker_threads: 1

v0.2.1 (Week 3) - Stability

Type: Patch Release

Fixes

  • 🐛 Fix sidecar worker memory leak
  • 🐛 Optimize event queue performance
  • 📊 Add sidecar analysis metrics (success rate, latency)

v0.3.0 (Week 4) - Smart Analysis

Type: Minor Release

Enhanced Intelligence

  • 🧠 Context-Aware Filtering
    • Command history analysis (consecutive failures → higher priority)
    • User behavior tracking (immediate retry → lower priority)
    • Time-based adjustment (late night → lower priority)

Configurable Strategies

sidecar:
  analysis_mode: smart  # smart | aggressive | minimal
  notification_style: indicator  # indicator | toast | silent

  # Smart mode rules
  smart_rules:
    ignore_commands: [grep, diff, test, ssh]
    ignore_exit_codes: [130]  # Ctrl-C
    benign_stderr_patterns:
      - "^Warning:"
      - "^Note:"

New Commands

  • :ai clear - Clear all suggestions
  • :ai stats - Show analysis statistics

Improvements

  • 95% reduction in false positives
  • < 50ms command execution overhead
  • Configurable notification styles

🛠️ Phase 2: Core Capabilities (Week 5-8)

v0.3.1 (Week 5) - Refinement

Type: Patch Release

Fixes

  • 🐛 Fix edge cases in smart filtering
  • 🐛 Optimize suggestion storage performance

v0.4.0 (Week 6) - Plan Mode & Tools

Type: Minor Release

Plan Mode (Inspired by Claude Code)

  • 🎯 PlanAgent: Task decomposition expert
  • 📋 User approval workflow before execution
  • 💾 Plan persistence to .aish/plans/

Usage

aish> ;deploy the application to production
[AI creates plan]
┌─ Deployment Plan ─────────────────────────────────────┐
│ 1. Run test suite                                     │
│ 2. Build Docker image                                 │
│ 3. Push to registry                                   │
│ 4. Update Kubernetes deployment                       │
│                                                        │
│ Approve? [y/N]:                                       │
└───────────────────────────────────────────────────────┘

Enhanced Tool Suite

  • 🔍 WebSearchTool: DuckDuckGo integration
  • 🔧 GitTool: Git operations wrapper (status, diff, commit, push)
  • 🧠 CodeAnalysisTool: AST-based code analysis (tree-sitter)

New Commands

  • :plan - Enter plan mode
  • :plan show - View current plan
  • :plan approve - Approve and execute plan

v0.4.1 (Week 7) - Polish

Type: Patch Release

Fixes

  • 🐛 Fix plan mode edge cases
  • 🐛 Optimize WebSearch result parsing
  • 📝 Add plan mode documentation

v0.5.0 (Week 8) - Multi-Agent System

Type: Minor Release

Agent Ecosystem

  • 🤖 CodeReviewAgent: Static analysis + best practices
  • 🐛 DebugAgent: Log analysis + root cause identification
  • 🔍 ResearchAgent: Web + documentation search
  • 🎭 AgentOrchestrator: Parallel/sequential agent coordination

Agent Architecture

# Example: Parallel agent execution
aish> ;review this PR and check for security issues

[Spawning agents in parallel]
├─ CodeReviewAgent: Analyzing code quality...
├─ SecurityAgent: Scanning for vulnerabilities...
└─ TestAgent: Checking test coverage...

[Results aggregated and presented]

Smart Context Management

  • 📊 Priority-based message ranking
  • 🗜️ Auto-summarization of long conversations (using small model)
  • 🧠 Cross-session knowledge base (optional vector search)

Configuration

agents:
  enabled: true
  max_parallel: 3
  context_window: 8000
  auto_summarize: true

🎨 Phase 3: User Experience & Ecosystem (Week 9-12)

v0.5.1 (Week 9) - Stability

Type: Patch Release

Fixes

  • 🐛 Fix agent parallel execution race conditions
  • 🐛 Optimize context compression algorithm
  • 📊 Add agent performance metrics

v0.6.0 (Week 10) - Rich UI & Tasks

Type: Minor Release

Rich UI Enhancements

  • 🎨 Real-time progress bars for agent execution
  • 🌳 Task tree visualization
  • 🔍 Interactive confirmation panels with diff preview

Task Management System (Inspired by Claude Code)

  • 📋 Built-in task tracking (TaskCreate, TaskUpdate, TaskList)
  • 🔗 Task dependencies and priorities
  • 💾 Task persistence and recovery

Usage

aish> :task list
┌─ Active Tasks ────────────────────────────────────────┐
│ [1] ⏳ Implement user authentication                  │
│     ├─ [2] ✅ Set up database schema                  │
│     ├─ [3] 🔄 Create login endpoint                   │
│     └─ [4] ⏸️  Add JWT token validation               │
└───────────────────────────────────────────────────────┘

aish> :task show 3
[Detailed task view with progress and blockers]

New Commands

  • :task create - Create new task
  • :task list - List all tasks
  • :task show <id> - View task details
  • :task complete <id> - Mark task as complete

v0.6.1 (Week 11) - Optimization

Type: Patch Release

Fixes

  • 🐛 Fix task manager memory usage
  • 🐛 Optimize Rich UI rendering performance
  • 📝 Add task management documentation

v0.7.0 (Week 12) - Agent SDK & MCP

Type: Minor Release

Agent SDK

  • 🔌 Standardized agent development interface
  • 🏗️ Agent template generator (aish create-agent)
  • 🌐 Agent marketplace (community sharing)

Agent Development Example

# Create new agent
aish create-agent --name my-agent --type diagnostic

# Generated structure
~/.config/aish/agents/my-agent/
├── agent.py          # Agent implementation
├── config.yaml       # Agent configuration
├── README.md         # Documentation
└── tests/            # Unit tests

MCP Protocol Support

  • 🔗 Compatible with Claude Desktop MCP servers
  • 🔌 Built-in MCP client for external services
  • 📡 Bidirectional communication with MCP ecosystem

Observability Dashboard (Optional)

  • 📊 Web UI (FastAPI-based)
  • 📈 Real-time session monitoring
  • 💰 Token usage statistics
  • ⚡ Agent performance analytics

Configuration

mcp:
  enabled: true
  servers:
    - name: filesystem
      command: npx
      args: [-y, @modelcontextprotocol/server-filesystem, /tmp]
    - name: github
      command: npx
      args: [-y, @modelcontextprotocol/server-github]
      env:
        GITHUB_TOKEN: ${GITHUB_TOKEN}

observability:
  enabled: false  # Optional web dashboard
  port: 8080

📊 Feature Comparison Matrix

Feature v0.1.0 (Current) v0.2.0 v0.4.0 v0.6.0 v0.7.0
Async AI Analysis ❌ ✅ ✅ ✅ ✅
Smart Filtering ❌ ⚠️ Basic ✅ Advanced ✅ ✅
Plan Mode ❌ ❌ ✅ ✅ ✅
Multi-Agent ⚠️ 1 Agent ⚠️ 1 Agent ⚠️ 1 Agent ✅ 4+ Agents ✅
Task Management ❌ ❌ ❌ ✅ ✅
Agent SDK ❌ ❌ ❌ ❌ ✅
MCP Protocol ❌ ❌ ❌ ❌ ✅
Rich UI ⚠️ Basic ⚠️ Basic ⚠️ Basic ✅ ✅

🎯 Competitive Positioning

vs Claude Code

Aspect AISH v0.7.0 Claude Code
Open Source ✅ Apache 2.0 ❌ Proprietary
Local Models ✅ Full support ❌ Limited
Multi-Provider ✅ LiteLLM ⚠️ Mainly Claude
PTY Support ✅ Full ⚠️ Limited
Ops Focus ✅ System diagnostics ⚠️ Dev-focused
Privacy ✅ Local-first ⚠️ Cloud-based
Cost ✅ Free ❌ Subscription
Async AI ✅ Non-blocking ⚠️ Blocking

Unique Value Propositions

  1. Non-Intrusive AI: Background analysis never blocks user workflow
  2. Ops-Native: Built for system administration and troubleshooting
  3. Privacy-First: Local model support with no data leaving your machine
  4. Community-Driven: Open source with extensible agent ecosystem
  5. Enterprise-Ready: Sandbox, audit logs, and fine-grained permissions

🚨 Risk Management

Risk Impact Mitigation Status
Async complexity High Extensive testing + fallback mode Week 1-2
Smart filter accuracy Medium Configurable rules + user feedback Week 3-4
Worker resource usage Medium Queue limits + auto-throttling Week 2-3
User adoption Low Progressive migration + docs Ongoing
Agent coordination Medium Timeout limits + error recovery Week 7-8

📈 Success Metrics

Technical KPIs

  • ✅ Command execution latency < 50ms (P95)
  • ✅ AI analysis false positive rate < 5%
  • ✅ Sidecar worker memory < 50MB
  • ✅ Agent response time < 2s (P95)
  • ✅ Test coverage > 80%

User KPIs

  • 📈 Daily active users growth > 20% MoM
  • ⭐ User satisfaction > 4.5/5
  • 🤝 Community skills > 50
  • 🏢 Enterprise deployments > 10

Ecosystem KPIs

  • 🔌 Community agents > 20
  • 📦 MCP integrations > 5
  • 📝 Documentation completeness > 90%

🤝 Contributing

We welcome contributions! Priority areas:

Week 1-4 (Phase 1)

  • 🔧 Sidecar architecture implementation
  • 🧪 Smart filtering rule development
  • 📝 Migration guide documentation

Week 5-8 (Phase 2)

  • 🤖 New agent implementations
  • 🔍 Tool integrations (web search, code analysis)
  • 🧠 Context management optimization

Week 9-12 (Phase 3)

  • 🎨 UI/UX improvements
  • 🔌 Agent SDK development
  • 📊 Observability dashboard

See CONTRIBUTING.md for detailed guidelines.

📚 Resources

📝 Changelog

Upcoming

  • See individual release sections above

v0.1.0 (Current)

  • ✅ Full PTY support
  • ✅ Multi-model support (LiteLLM)
  • ✅ Basic security risk assessment
  • ✅ Skills hot-reload system
  • ✅ ReAct diagnostic agent
  • ✅ Session persistence (SQLite)
  • ✅ Output offload mechanism
  • ✅ i18n support

Last Updated: 2026-03-06 Roadmap Version: 2.0 Status: 🟢 Active Development

Clone this wiki locally