Melra

Melra is an open-source local runtime that checks AI agent tool calls, gates risky actions on approval, and verifies results.

Melra is an open-source local execution runtime for AI agent tool calls. It checks each tool call against a defined policy, requests approval for risky actions, verifies declared evidence to ensure success, and records a redacted receipt of the outcome. The runtime operates between the agent and its actions, dispatching mutations at most once and parking uncertain outcomes for reconciliation instead of blind replay. It supports numerous agent clients through the Model Context Protocol, including Claude Code, Cursor, VS Code, and others. Developers can set up the runtime using the CLI tool or integrate it via TypeScript and Python SDKs.

Melra at a glance

Best for: Software engineer
Runs on: Web, macOS, Windows, Linux, CLI, API, VS Code, JetBrains
Pricing: Free
Released: October 2026
Category: AI agent tool
Website: github.com/xagi-lab/melra

Key features

Checks tool calls against local security policies
Requests manual approval for risky actions
Verifies declared evidence to confirm success
Records redacted, hash-linked execution receipts
Prevents blind replays after timeouts or crashes
Integrates with Model Context Protocol clients

Use it for

Authorizing and gating AI agent file edits
Controlling command-line tool execution
Auditing AI agent actions with secure receipts
Managing safe tool call retries and recoveries

Alternatives to Melra

iFixAi — Audits agent behavior independently without building manual eval harnesses.
Rakazo — Complete ownership of data, models, and execution environment.
Composio — Equips AI agents with tools, managed user auth, and sandboxes across 1,500+ apps
MyAgent — It features built-in permission controls, long-term memory, and skill self-evolution.
OpenAPPA — Tracks data flows algebraically to prevent prompt injection without breaking agents.
Kora Doctor — Surfaces waste across tool usage, context growth, and repeated inference without an account.
Yana AI — Governs AI model actions with a local control plane and human approvals.
APIStack — Connect AI agents to over 100 tools using one API key or a single MCP endpoint.
BrowserOS — Runs entirely locally and lets AI agents use real user logins to complete tasks.
Permaura — Enforces user-defined rules and device-signed approvals for AI actions across connected apps.
Kavela — A reviewer agent checks every task against set rules before it reaches the team.
Bees — Runs AI agents locally on your computer while keeping files, passwords, and sessions private.

All Melra alternatives