Roadmap

ttal is actively developed. Here's what's on the horizon.

v1.x — Current

The foundation: multi-agent orchestration over Telegram with Claude Code and Codex CLI support.

• Agent lifecycle — spawn, monitor, and clean up coding workers
• Telegram bridge — bidirectional messaging between humans and agents
• PR workflow — create, review, merge PRs from your phone
• Task routing — design, research, test, execute pipeline
• Voice — text-to-speech via Kokoro for agent responses
• Multi-runtime — Claude Code, Codex CLI
• Starter templates — ttal init scaffolds for new teams

v1.x target: multi-runtime GA

The v1.x goal is making Codex CLI support production-ready alongside Claude Code — so both runtimes are fully supported for spawning, messaging, PR workflows, and review.

In progress:

• Codex CLI runtime parity (spawn, review, yolo modes)
• Filesystem-based agent discovery (replacing DB-stored agents)
• ttal init interactive scaffold picker
• Configurable prompts for all task routing

v2.0 — Matrix Protocol

Replace Telegram with Matrix as the primary communication layer.

Why Matrix:

• Unlimited bots — Telegram limits bot interactions; Matrix has no such constraints
• Rich markdown — full markdown rendering in messages (code blocks, tables, formatting)
• Self-hosted — run your own Synapse server, full data sovereignty
• E2EE — end-to-end encryption for agent communications
• Rooms as channels — dedicated rooms per agent, per project, per workflow
• Bridges — Matrix bridges to Telegram, Slack, Discord for gradual migration

Planned changes:

• Matrix adapter alongside existing Telegram adapter
• Room-per-agent architecture
• Threaded conversations for task context
• File sharing (plans, research docs, screenshots) directly in chat

v2.x — Sandboxed Execution

Worker execution sandboxing via temenos (seatbelt on macOS, bwrap on Linux).

Shipped:

• Isolation — workers can't affect host system or other workers
• Security — untrusted code execution in a safe boundary
• Resource limits — cap CPU, memory, disk per worker

Approach: OS-native sandboxing (seatbelt/bwrap) via temenos — no containers needed.