Tetora v2.1.0 — Massive Code Consolidation + Workflow Engine

Tetora v2.1.0 is a significant release. Two themes define this version: a major codebase restructure and new features shipping.

For users: a more stable runtime, faster iteration cycles, and the long-awaited Workflow Engine and Template Marketplace. For developers: this is Tetora’s transition from rapid prototype to long-term maintainable product.

TL;DR: Root source files consolidated from 28 to 9, test files from 111 to 22, the entire repo compressed from 256+ files to a lean structure — while adding the Workflow Engine, Template Marketplace, and multiple Dashboard improvements.

Codebase Consolidation: 256 Files → Lean Core

The consolidation happened across multiple refactoring rounds. Here are the final numbers:

The 9 root files are split by domain:

• main.go — Entry point, command routing, startup sequence
• http.go — HTTP server, API routing, Dashboard handlers
• discord.go — Discord gateway, message handling, terminal bridge
• dispatch.go — Task dispatch, TaskBoard, concurrency control
• workflow.go — Workflow Engine, DAG execution, step management
• wire.go — Cross-module wiring, initialization, dependency injection
• tool.go — Tool system, MCP integration, capabilities management
• signal_unix.go / signal_windows.go — Platform-specific signal handling

Substantial business logic has been migrated to internal/ sub-packages: internal/cron, internal/dispatch, internal/workflow, internal/taskboard, internal/reflection, and more. The root layer now holds only a thin coordination layer.

Why Does This Matter?

Previously, the root layer had 100+ files with functionality scattered everywhere — new contributors had to spend significant time just finding the right file. After consolidation:

• Easier to maintain — you know exactly where to look when changing a feature
• Faster onboarding — 9 entry points instead of 28
• Clearer IDE navigation — go to definition no longer gets lost in a sea of files
• Faster builds — fewer unnecessary package boundaries and import chains

Workflow Engine

The Workflow Engine is the flagship new feature of v2.1.0. Define multi-step AI workflows in YAML; Tetora handles execution, error recovery, and state tracking.

DAG-Based Pipeline

Workflows are defined as directed acyclic graphs (DAGs), supporting:

• Conditional branching — route execution based on the output of prior steps
• Parallel steps — steps with no dependencies run concurrently, reducing total wall time
• Retry mechanism — automatic retries on step failure, configurable count and backoff strategy

name: content-pipeline
steps:
  - id: research
    agent: hisui
    prompt: "Research the topic: {{input.topic}}"
  - id: draft
    agent: kokuyou
    depends_on: [research]
    prompt: "Write a draft based on the research"
  - id: review
    agent: ruri
    depends_on: [draft]
    condition: "{{draft.word_count}} > 500"

Dynamic Model Routing

The Workflow Engine automatically selects the model based on task complexity:

• Simple formatting, summarization → Haiku (fast, low cost)
• General reasoning, writing → Sonnet (default)
• Complex analysis, multi-step planning → Opus (maximum capability)

You can also specify the model explicitly in YAML, or let the router decide based on prompt length and keywords.

Dashboard DAG Visualization

Running workflows are rendered as a node graph in the Dashboard: completed steps in green, running steps with a purple animation, waiting steps in gray, failed steps in red. Real-time pipeline progress at a glance — no need to tail logs.

Template Marketplace

The Template Marketplace enables sharing, browsing, and one-click importing of workflow templates. It’s Tetora’s first step from a personal tool to a community ecosystem.

Store Tab

A new Store tab in the Dashboard provides:

• Category browsing — filter by domain (Marketing, Engineering, Finance, Research, etc.)
• Full-text search — search across template names and descriptions
• Featured section — officially curated, high-quality templates
• One-click import — click to import directly into your local workspace

Capabilities Tab

A new Capabilities tab gives a consolidated view of everything your Tetora instance can do:

• Tools — available MCP tool list
• Skills — defined Skill commands
• Workflows — local Workflow templates
• Templates — Agent prompt templates

CLI Import / Export

Not just UI — full CLI support too:

tetora workflow export my-pipeline   # Export as shareable YAML
tetora workflow create from-store    # Import a template from the Store
tetora workflow list                  # List all local workflows

Exported YAML can be pasted directly into a GitHub Gist or the Tetora Store to share with the community.

TaskBoard & Dispatch Improvements

Multiple important improvements to the TaskBoard and Dispatch layer improve stability and observability for multi-agent parallel workloads.

Configurable Parallel Slots + Slot Pressure

You can now specify the maximum number of parallel slots and a slot pressure threshold in the config. When system load exceeds the threshold, new tasks are queued rather than force-inserted, preventing agents from competing for resources:

{
  "dispatch": {
    "maxSlots": 4,
    "slotPressureThreshold": 0.8
  }
}

Partial-Done Status

Long-running tasks now support a partial-done intermediate state. Agents can report progress after completing portions of work; the TaskBoard displays a completion percentage so you know the task is advancing rather than stalled.

Worktree Data Protection

When multiple agents use Git worktrees for parallel development, explicit data isolation is now enforced. Each agent’s working directory is independent — no accidental overwrites or merge conflicts polluting another agent’s state.

GitLab MR Support

In addition to GitHub PRs, GitLab Merge Request workflows are now supported. The tetora pr create command auto-detects the remote type and calls either the GitHub CLI or GitLab CLI accordingly to open the MR.

Install / Upgrade

Fresh Install

curl -fsSL https://tetora.dev/install.sh | bash

Single binary, zero external dependencies. macOS, Linux, and Windows all supported.

Upgrading from a Prior Version

tetora upgrade

Automatically downloads the latest release, replaces the binary, and restarts the daemon. Tasks running at upgrade time are not interrupted.

Tip: Before upgrading, confirm there are no long-running workflows in progress. Run tetora status to check active tasks.

What’s Next: v2.2 Roadmap

With v2.1.0 shipped, development focus shifts to two theme modules for v2.2:

Financial Module

Personal and small business finance automation: income/expense tracking, report generation, budget monitoring. Integration with common accounting APIs (freee, Money Forward, etc.) is planned.

Nutrition Module

Health and diet tracking: meal logging, nutritional analysis, goal setting. Claude acts as a nutrition advisor, offering personalized recommendations based on your eating habits.

Both modules will be published to the Store as Workflow templates — import and use immediately, no configuration from scratch required.

Upgrade to v2.1.0 Now

Single binary, zero dependencies. macOS / Linux / Windows.

tetora upgrade

View Release Notes on GitHub