Tessera

No-code multi-agent AI orchestration for full project generation.

Like mosaic tiles coming together to form a complete picture, Tessera coordinates specialized AI agents to build entire software projects from scratch.

What is Tessera?

Tessera is a multi-agent AI orchestration framework for automated project generation.

Features: - Multi-agent coordination (supervisor, interviewer, specialist agents) - Multi-provider LLM support (OpenAI, Vertex AI, 100+ via LiteLLM) - Persistent memory system for agent learning - Workspace management with sandboxing - Real token tracking and cost calculation - Complete local observability (OpenTelemetry + SQLite) - Extensible tool system with MCP support

No coding required - define agents with markdown prompts.

Key Features

🎯 No-Code Operation

Define agents using markdown system prompts. No Python coding required.

🤝 Multi-Agent Coordination

Supervisor: Orchestrates task decomposition and delegation
Interviewer: Evaluates agents and gathers requirements
Specialists: Code reviewers, testers, security experts, and more

🚀 Parallel Execution

Run multiple agents concurrently with intelligent coordination and conflict resolution.

🔍 Full Observability

OpenTelemetry tracing (local JSONL files)
SQLite metrics (task history, costs, performance)
Real-time cost tracking
Agent performance analytics

🛠️ Extensible Tools

Built-in: File operations, Git, web search, code execution
Plugins: Add custom tools via Python files
MCP: Integrate Model Context Protocol servers

💰 Cost Management

Automatic cost calculation for 100+ LLM models
Configurable budget limits (daily, per-task, per-agent)
Cost threshold approvals

🧠 Persistent Memory

Long-term conversation history across sessions
Agent learnings and decisions storage
Vector-based semantic search
SQLite-backed persistence

📁 Workspace Management

Project isolation and tracking
Sandboxing with resource limits
Filesystem protection
Archive and restore capabilities

🔒 Security First

Configurable sandboxing (Docker, Podman, uv)
Risk-based approval gates
File locking for conflict prevention

📊 Intelligent Agent Selection

Performance-based routing
Adaptive learning from failures
Interview-driven capability assessment

Quick Start

Installation

# Option 1: Run directly with uvx
uvx tessera-agents init

# Option 2: Install globally, then use 'tessera' command
uv tool install tessera-agents
tessera init

This creates: - ~/.config/tessera/config.yaml - Your configuration - ~/.config/tessera/prompts/ - Agent system prompts

Note: Examples below use tessera commands (assumes global install). With uvx, use uvx tessera-agents <command>.

First Project

tessera

The interactive wizard will: 1. Ask about your project 2. Interview you for requirements 3. Generate a comprehensive plan 4. Execute with multiple agents in parallel 5. Deliver a complete, tested, documented project

Example

# Interactive mode - prompts for task description
$ tessera

# Direct task execution
$ tessera "Build a FastAPI backend with user authentication"

# Run without executing (plan only)
$ tessera --dry-run "Deploy application"

Architecture

Tessera uses:

LangGraph: Complex multi-agent workflow orchestration
Pydantic AI: Type-safe agent definitions
LiteLLM: Unified interface to 100+ LLM providers
OpenTelemetry: Vendor-neutral observability
SQLite: Local metrics and state persistence