JSPM

tycono

0.3.45-beta.3
  • ESM via JSPM
  • ES Module Entrypoint
  • Export Map
  • Keywords
  • License
  • Repository URL
  • TypeScript Types
  • README
  • Created
  • Published
  • Downloads 2945
  • Score
    100M100P100Q127412F
  • License MIT

Your AI team for solo builders. Run a full company with AI agents.

Package Exports

    This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (tycono) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

    Readme

    tycono

    AI Team Orchestration Engine

    Infrastructure-as-Code defines servers with a single YAML file. Company-as-Code defines an entire organization in code — and AI runs it.

    npm version server version license

    Real-World Results

    A crypto trading research team ran 13.5 hours of autonomous AI research with zero human intervention:

    Metric Value
    Duration 13.5 hours, 0 human input
    Sessions 174 agent sessions
    Hypotheses tested 29
    Cost $54 total ($2.5/hypothesis)
    Critic catches Prevented false-positive from reaching production

    Cost comparison: Solo agent $50/task vs Tycono $4/wave (down from $1,342 during beta).

    Philosophy

    Company as Code

    Define organizational structure, roles, authorities, and knowledge scopes in plain files. AI agents operate within this structure autonomously.

    # role.yaml
id: engineer
name: "Alex"
level: member
reports_to: cto
model: claude-sonnet-4

authority:
  autonomous:
    - "Code implementation"
    - "Bug fixes"
  requires_approval:
    - "Architecture changes"

knowledge_scope:
  readable: ["projects/", "architecture/"]
  writable: ["projects/*/technical/"]

    One file defines who the agent is, what they can do, what they know, and who they report to.

    Vision: tycono deploy company.yaml — deploy a company with one command.

    Knowledge That Compounds

    Every AI tool resets when the session ends. Tycono doesn't.

    The Agentic Knowledge Base (AKB) is a file-based knowledge system where insights persist, compound, and cross-link over time.

    1. Pre-K    Read existing knowledge before starting
2. Task     Define work from what the org already knows
3. Execute  Do the actual work
4. Post-K   Extract insights, cross-link, register in Hub
5. Next     Update status, identify next work

    Session 50 is dramatically smarter than Session 1.

    Multi-Agent Supervision

    Not a black box. You watch the entire hierarchy work in real time.

    dispatch   CEO assigns CTO. CTO assigns Engineer.
watch      Heartbeat monitors progress in real time.
react      If direction drifts, supervisor sends amend.
repeat     Until the wave is complete.

    Authority is enforced — engineers can't make CEO decisions, PMs can't merge code.

    Key Features (v0.1.0)

    Auto-Amend — 70% cost reduction

    When a subordinate needs follow-up work, Tycono automatically continues the existing session instead of creating an expensive new one. A Haiku classifier decides whether each dispatch is a follow-up (amend) or genuinely new work (dispatch).

    Dispatch Protocol

    Fire-and-forget async dispatch with real-time supervision. dispatch:start and dispatch:done events flow between parent and child sessions. Supervisors use heartbeat watch for progress monitoring.

    Handoff Summary

    When the same role is re-dispatched, previous session results are automatically injected into the new context. No starting from scratch.

    Prompt Caching

    System prompt sections are tagged as cacheable/dynamic. Static sections (role definition, skills, org structure) get cache_control for up to 90% input cost reduction on the Direct API runner.

    Active Wave Guard

    Server shutdown checks for running waves before killing. Non-interactive environments block concurrent wave launches. Prevents data loss from accidental server kills.

    Critic Role

    A Devil's Advocate role that challenges team conclusions. In production, a Critic caught a false-positive test result that would have been deployed, and blocked a diminishing-returns research loop.

    Surfaces

    One engine, three interfaces.

    Plugin (Claude Code)

    Install the plugin and run AI teams directly inside Claude Code.

    claude plugin install tycono
/tycono "Build a landing page"
/tycono --agency gamedev "Build a tower defense"

    TUI (Terminal)

    Terminal-native dashboard with multi-wave workspaces, real-time agent streams, and org tree visualization.

    npx tycono
    ┌── W1 Build the API ──────┬── Stream  Docs  Info ──────────────┐
│  3 sessions               │  cto     → dispatch engineer       │
│  ── Org Tree ──           │  engineer → Write src/api/routes.ts│
│  ● CEO                    │  engineer  Write src/api/types.ts  │
│  ├─ ● CTO               │  qa       ▶ Running test suite...   │
│  │  ├─ ● engineer         │                                     │
│  │  └─ ● qa               │                                     │
│  └─ ○ CBO                │                                     │
└───────────────────────────┴─────────────────────────────────────┘

    Pixel (Web Dashboard)

    Isometric pixel office visualization. Watch your AI team walk around, sit at desks, and collaborate visually.

    Coming Soon at tycono.ai

    Pixel office visualization

    Agencies

    Pre-built team configurations for common workflows.

    /tycono --agency gamedev "Build a browser game"
/tycono --agency saas "Build an MVP for our SaaS"
/tycono --agency research "Analyze competitor landscape"

    Browse all agencies at tycono.ai/agencies

    Quick Start

    mkdir my-company && cd my-company
npx tycono

    A setup wizard guides you through naming your company, choosing a team template, and starting work.

    Requirements

    Project Structure

    This is a monorepo:

    tycono/
├── src/
│   ├── api/              ← Core Engine (tycono server)
│   ├── tui/              ← TUI (terminal interface)
│   └── web/              ← Pixel (original web UI)
├── packages/
│   ├── server/           ← npm: tycono-server
│   ├── plugin/           ← Claude Code Plugin
│   └── web/              ← Agency Hub (tycono.ai)
├── .github/assets/       ← Screenshots
└── README.md

    Your Company Structure

    When you run npx tycono, it scaffolds:

    your-company/
├── CLAUDE.md             ← AI operating rules
├── roles/                ← AI role definitions (role.yaml + skills)
├── projects/             ← Product specs, PRDs, and tasks
├── architecture/         ← Technical decisions and designs
├── knowledge/            ← Domain knowledge (compounds over time)
└── .tycono/              ← Config and preferences

    CLI

    npx tycono                # Start TUI (default)
npx tycono ./my-company   # Start with specific directory
npx tycono --classic      # Pixel office web UI
npx tycono --attach       # Connect to running API server
npx tycono --help         # Show help

    TUI Commands

    Type naturally to talk to your AI team.

/new [text]       Create new wave
/waves            List all waves
/focus <n>        Switch to wave n
/agents           Wave → Role → Session tree
/sessions         Sessions + ports
/kill <id>        Kill a session
/docs             Files created in this wave
/read <path>      Preview file content
/open <path>      Open in $EDITOR
Tab               Panel Mode (org tree + stream + docs)
/quit             Exit

    Screenshots

    Wave center
    Multi-wave workspace with real-time agent streams

    Knowledge graph
    Knowledge graph visualization

    Pro view
    Pro view with detailed agent activity

    Development

    git clone https://github.com/seongsu-kang/tycono.git
cd tycono
npm install
cd src/api && npm install && cd ../..
cd src/web && npm install && cd ../..

# Dev mode (hot reload)
npm run dev