OpenTag Review 2026 — Open-Source @Agent Mentions for Slack and GitHub

What Is OpenTag?

OpenTag is an open-source, local-first agent gateway that lets you mention a coding agent directly from Slack or GitHub. When someone writes @opentag investigate this error in a thread, OpenTag routes the request to Codex or Claude Code running on your infrastructure and posts the results back to the same thread.

Developed by Amplify HQ (the team behind Codex integrations), OpenTag hit 825 GitHub stars within two weeks of its June 24, 2026 launch. It’s positioned as a more open, self-hosted alternative to proprietary agent platforms.

Key stats:

• 825★ GitHub stars, MIT License
• Launched June 24, 2026
• Supports Slack, GitHub, GitLab, Lark/Feishu, Telegram, Discord
• npm package: @opentag/cli

How OpenTag Works

OpenTag treats the thread where a request starts as the approval surface for agent-proposed mutations. The flow is:

1. Mention — Someone writes @opentag <task> in Slack or GitHub
2. Context Curation — OpenTag gathers the surrounding thread context: messages, files, code snippets
3. Permission Check — The dispatcher checks what the executor is capable of and what permissions apply
4. Agent Execution — Codex or Claude Code runs locally with the curated context
5. Receipt Generation — OpenTag renders a compact action receipt showing what will change
6. Approval — The team member approves or rejects the proposal
7. Apply — If approved, the change is applied (e.g., a PR is opened)
8. Audit Trail — Every run records a full agent work ledger: source event, context snapshot, executor capability, produced artifacts, callback delivery, and final outcome

Source-Thread Action Receipts

The receipt system is OpenTag’s most thoughtful feature. When an agent suggests a change, instead of dropping raw code blocks into the chat, OpenTag renders a structured receipt:

• What will change — files affected, diff summary
• Ready to apply — whether the dispatcher can execute the action
• Decision buttons — Apply, Review, or Dismiss

The “Apply” button appears only when the dispatcher confirms a configured adapter can execute the action. Otherwise, it shows “setup needed” or “attention required.”

Agent Work Ledger

Each run maintains a local audit trail accessible through:

opentag status --run <run_id>

The ledger captures:

• Source event (which thread, who mentioned it)
• Admission decision (was the request accepted?)
• Context packet snapshot (what was sent to the agent)
• Executor capability snapshot (what the agent was allowed to do)
• Produced artifacts (code changes, analysis, etc.)
• Callback delivery (was the result posted back?)
• Final outcome (applied, rejected, errored)

Quick Start

npm install -g @opentag/cli@latest
opentag setup

The setup wizard guides through:

1. Language preference for the CLI
2. Port configuration
3. Agent selection (Codex or Claude Code)
4. Local project association
5. Platform credential setup (Slack, GitHub, etc.)
6. Service mode (background daemon or terminal)

For automation:

opentag setup --service

This installs and starts the background service — LaunchAgent on macOS, systemd —user on Linux.

Multi-Platform Support

Platform	Status	Notes
Slack	✅	App registration required
GitHub	✅	Webhook configuration
GitLab	✅	API token setup
Lark/Feishu	✅	Bot configuration
Telegram	✅	Bot token
Discord	✅	Bot application

The broad platform support is unusual for an open-source tool in this space. Most competitors focus on one or two platforms.

Community Reception

The HN thread for OpenTag (trending on July 6, 2026) generated discussion around the “local-first” approach vs. cloud-based alternatives like GitHub Copilot Chat. The main points of discussion:

• Privacy advocates praised the local-first architecture — no agent data leaves the network
• Self-hosting enthusiasts liked the MIT License and transparent audit trail
• Scaling concerns were raised about the lack of a cloud broker — each developer needs to configure local infrastructure
• The receipt system was widely praised as the right UX for agent-in-the-loop workflows

A notable comment from the HN thread: “This is what I wanted GitHub Copilot Chat to be — a connector, not a walled garden.”

Use Cases

On-Call Triage: When a production error comes into a Slack thread, an engineer writes @opentag investigate this stack trace and gets a root cause analysis without leaving the incident channel.

Code Review Assistance: In a GitHub PR thread: @opentag review the security implications of this change — the agent analyzes the diff and posts findings as a thread reply.

CI/CD Integration: @opentag fix the failing test in CI — the agent reads the test output, diagnoses the failure, and proposes a fix as a PR.

Natural Language Tasking: @opentag add input validation to the signup form — the agent creates the implementation and returns a receipt showing the changes.

Comparison to Alternatives

	OpenTag	GitHub Copilot Chat	Custom Bot
Open source	✅ MIT	❌ Proprietary	✅
Local-first	✅	❌ Cloud	⚠️ Depends
Multi-platform	✅ 6 platforms	❌ GitHub only	⚠️ Custom build
Action receipts	✅ Structured UI	❌ Raw code	⚠️ Custom
Audit ledger	✅ Full trace	❌	⚠️ Custom
Setup effort	Medium (self-host)	Low (SaaS)	High
Agent choice	Codex, Claude Code	Copilot only	Any

Pricing

OpenTag is completely free and open-source under MIT License. You run it on your own infrastructure. There is no paid plan, no cloud tier, and no API keys to manage beyond your platform credentials.

Verdict

OpenTag solves a real problem well: bridging the gap between team collaboration platforms and AI coding agents. The local-first architecture, multi-platform support, and structured receipt system make it a strong choice for engineering teams that already use Claude Code or Codex.

The self-hosted nature and lack of a cloud option mean it’s best suited for teams with DevOps capabilities. But for those teams, OpenTag provides a polished, auditable, and flexible agent gateway that goes beyond what proprietary alternatives offer.

Rating: 8.0/10 — a well-designed tool for a clear use case. The self-hosted requirement limits its reach, but for teams that can manage it, OpenTag delivers a genuinely useful agent interaction model.

Note: The cover image is a screenshot of the OpenTag GitHub repository showing the README, star count, and supported platforms.