Collaborative AI development - A revolutionary context-first development environment where AI agents collaborate autonomously to build software.
TENEX represents a paradigm shift in software development. If LLMs have fundamentally changed how we write code, shouldn't our development environments evolve too? TENEX answers this by replacing the traditional text editor with a context-first environment where context, not code, becomes the primary building block.
At its core, TENEX is a sophisticated multi-agent coordination system built on the Nostr protocol. It enables autonomous AI agents to collaborate on complex software development tasks through intelligent routing, phase-based workflows, and continuous learning.
Unlike traditional AI assistants where you interact with a single entity, TENEX employs an intelligent routing pattern. The system automatically routes your requests to specialized agents best suited for each task, creating a seamless experience where the right expertise appears exactly when needed.
- Specialized Agents: Agents for planning, execution, and project management fetched from Nostr
- Dynamic Routing: Intelligent task delegation based on agent capabilities and context
- Parallel Execution: Multiple agents can work simultaneously on different aspects of your project
- Custom Agents: Extensible system allowing you to define domain-specific experts
Agents can define phases (for example: planning, execution, review) to structure their own todo lists and to pass phase instructions when delegating. Phases are optional and vary per agent.
- Agents capture and apply lessons from every interaction
- Cross-conversation knowledge sharing
- Self-improving behavior based on accumulated experience
- Persistent knowledge base stored on Nostr
- Comprehensive Tools: File operations, code analysis, shell execution, Git integration
- MCP Integration: Model Context Protocol support for dynamic tool loading
- Type-Safe: Comprehensive validation and error handling
- Composable: Tools can be combined for complex operations
- Decentralized by Design: No central server, peer-to-peer agent communication
- Cryptographic Identity: Each project maintains its own nsec for secure context
- Event Sourcing: Complete audit trail of all agent actions
- Resilient: Continues operating even with network disruptions
- Support for OpenAI, Anthropic, Google, OpenRouter, and more
- Intelligent model selection based on task requirements
- Automatic failover and rate limiting
- Cost optimization through smart routing
- A Rust toolchain (stable, via rustup)
- Git for version control integration
- An API key for at least one LLM provider (OpenAI, Anthropic, OpenRouter, …)
# Clone the repository
git clone https://github.com/tenex-chat/tenex
cd tenex
# Build the workspace
cargo build --releaseThe tenex binary is produced at target/release/tenex.
Before using TENEX, run the setup wizard to configure your identity, relays, and LLM provider credentials:
tenex onboardThen start the project supervisor, which subscribes to Nostr and spawns a runtime per project on inbound events:
tenex daemon-
Create a new project using the TENEX Web Client or iOS client
-
Example interaction:
You: Create a REST API for a todo application with authentication
TENEX: [System routes to Planner]
[Planner creates structured approach]
[Executor implements the API]
[Verification runs tests]
[Documentation is updated]
[Lessons are captured for future use]
- Architecture: Core principles, the Rust crate workspace, and module organization.
- crates/AGENTS.md: Modularization philosophy and the rules for adding or splitting crates.
- Module Inventory: Canonical map of every crate and its internal modules.
- Contributing: Development workflow, coding guidelines, and testing.
TENEX is a Rust workspace. The host CLI/supervisor lives in tenex/; every
other concern is a focused crate under crates/. See
MODULE_INVENTORY.md for the full map.
tenex/ # Host CLI + supervisor: onboarding, config, doctor, daemon, runtime, agent/mcp/cron commands
crates/
├── tenex-agent/ # One-shot agent runner: LLM loop + tools (NDJSON over stdio)
├── tenex-conversations/ # SQLite conversation store (messages, prompt history, completions, delegations)
├── tenex-context/ # Conversation-history → LLM-message projection
├── tenex-system-prompt/ # Deterministic system-prompt assembly
├── tenex-project/ # Read-side project view: membership, teams, signer
├── tenex-agent-registry/ # Global installed-agent JSON registry
├── tenex-llm-config/ # LLM/provider configuration resolver
├── tenex-mcp/ # Project-scoped MCP server runtime
├── tenex-rag/ # RAG storage + embeddings
├── tenex-summarizer/ # kind:513 conversation-metadata daemon
├── tenex-embedder/ # Conversation embedding daemon
├── tenex-scheduler/ # Cron / schedule daemon
├── tenex-intervention/ # Completion-timeout watcher daemon
├── tenex-identity/ # kind:0 profile resolver + cache
├── tenex-telegram/ # Telegram integration
├── tenex-protocol/ # TENEX intent vocabulary + Nostr channel encoding
├── tenex-whitelist/ # Local trust-set reader
├── tenex-supervision/ # Pure supervision heuristics (no I/O)
├── tenex-telemetry/ # OpenTelemetry / tracing bootstrap
└── tenex-accounting/ # LLM accounting store + embedded UI
We welcome contributions! Please read our Contributing Guide for a detailed overview of our development workflow, coding guidelines, and architectural principles.
# Build the workspace
cargo build
# Run the test suite
cargo test --workspace
# Lint (deny warnings)
cargo clippy --workspace --all-targets
# Format
cargo fmt --all
# Build optimized binaries
cargo build --releaseThe pre-commit hook runs cargo check --workspace.
Traditional IDEs optimize for code editing. TENEX optimizes for context management, recognizing that in the LLM era, maintaining and utilizing context effectively is more valuable than syntax highlighting.
The coordination system operates behind the scenes, presenting a simple conversational interface while managing sophisticated multi-agent collaboration underneath.
Mandatory verification and reflection phases ensure every task meets quality standards and contributes to the system's collective knowledge.
Built on Nostr from the ground up, not as an afterthought. This enables censorship-resistant, peer-to-peer agent networks with no single point of failure.
- Rapid Prototyping: Go from idea to working prototype through natural conversation
- Code Migration: Modernize legacy codebases with intelligent refactoring
- Documentation: Automatic generation and maintenance of technical documentation
- Testing: Comprehensive test generation and verification
- Learning: Agents that get better at your specific codebase over time
- Web-based interface improvements
- Multi-model ensemble execution
- Real-time collaborative editing
- Advanced debugging and profiling tools
MIT
- GitHub Issues: Report bugs or request features
- Nostr: Follow the project at
npub1tenex... - Documentation: Full documentation
Ready to experience the future of software development? Create your first project using the TENEX Web Client and let your AI agents handle the rest.
"The best code is the code you don't have to write. The second best is code written by agents who learn from every line they produce." - TENEX Philosophy