ClawMetry is an open-source observability tool for real-time monitoring of multiple AI agent runtimes. Below are 8 llm eval & observability apps with similar functionality to ClawMetry, matched by what each product actually does — not ranked or scored. Explore each to find the closest fit for your use case.
ClawMetry provides a live monitoring solution for AI agent activity, offering users immediate visibility into the operations of their agents and sub-agents. The platform is designed to help users track every agent, session, and tool call as they happen, giving a real-time window into ongoing processes. ClawMetry emphasizes monitoring token usage by tying every token to the specific task that consumed it, enabling users to keep close tabs on spending and resource allocation. A notable feature of ClawMetry is its ability to detect and alert users to runaway loops, flagging these before they can result in excessive costs. The service supports integration with 14 different runtimes, including OpenClaw, NVIDIA NemoClaw, NanoClaw, PicoClaw, Claude Code, Codex, Cursor, Aider, Goose, OpenCode, Qwen, Hermes, Pi, and Deep Agents, consolidating monitoring across these environments into a single dashboard. Users can connect their agents to ClawMetry Cloud by installing the platform via a provided script and connecting with an API key, which works across multiple machines. ClawMetry is accessible via a web dashboard and requires users to sign in using GitHub, Google, or email verification. The platform offers end-to-end encryption and allows users to start for free, with additional information about plans and pricing available through its site. The tool is positioned for those managing AI agents who need real-time oversight of activity and costs across diverse agent frameworks.
Claw Lens provides a local dashboard designed specifically for users of OpenClaw AI agents, offering a comprehensive suite of observability, cost analytics, security auditing, and debugging tools. The platform addresses the challenge of understanding and monitoring the internal operations of AI agents, making otherwise opaque processes—such as tool calls, retries, and token usage—visible in real time. It is intended for those who run OpenClaw agents locally and want detailed insight into their behavior, performance, and security. The dashboard features a live monitor that streams agent activity as it happens, showing each agent’s status, tool calls, and potential issues like looping, all delivered over WebSocket. Cost and token analytics allow users to track spending by day, model, or agent, analyze cache hit rates, and break down costs per session. Sessions are presented in a filterable table, with the ability to drill down into detailed metrics such as model usage, context pressure, cache activity, errors, and latency. The session timeline displays a Gantt chart of every LLM call and tool invocation, helping users identify where time is spent and spot performance bottlenecks. Additional features include an agent overview with health status, error history, and activity trends; a security audit that automatically scans for credential leaks, prompt injection attempts, dangerous commands, and data exfiltration, providing risk scoring and per-agent security profiles; and a context breakdown visualization to track how prompts, history, and tool results fill the context window during API calls. The tool also offers cron job detection, tool profiling for latency and failures, deep turn detection for identifying looping behavior, and cache trace visualization for debugging cache misses. Claw Lens operates entirely on the user’s machine, requiring no account, API key, or external service. It is launched via a single command and auto-detects the OpenClaw home directory to read agent logs locally, ensuring all data remains private. js 18 or higher.
Clawwatcher is a web-based tool for monitoring OpenClaw AI agents, offering real-time dashboards for token usage, budget limits, and spending alerts. It helps AI operations teams control costs and prevent budget overruns when deploying AI agents.
Clawbotomy focuses on behavioral intelligence for AI models, addressing the challenge of understanding how models behave under varying and stressful conditions. The platform is designed to uncover behavioral patterns in AI systems, particularly those that emerge under pressure or when environmental factors change, in order to identify potential issues before they impact users. One of the core features of Clawbotomy is its use of behavioral probes. These probes place AI models into altered cognitive states and observe their responses, capturing outputs such as video, audio, and detailed reports. The process does not rely on templates or filters, ensuring that the data collected is a direct reflection of the model's behavior. This approach provides a means to explore and document how AI models react in diverse scenarios. Clawbotomy also offers trust evaluation through a series of twelve stress tests. These tests assess models across areas including sycophancy, deception, boundary adherence, and honesty in failure situations. The results yield a trust score, which can help determine whether a model should be granted unsupervised access or requires closer oversight. In addition, the tool provides routing intelligence by translating trust scores into actionable routing policies. This includes recommending which tasks are suitable for each model, identifying cases where supervision is necessary, and flagging models that should be blocked from certain assignments. Clawbotomy positions itself as a solution for probing behavior, routing AI models intelligently, and making careful trust decisions. evidence_sufficient": true}
Openclaw Command Center is a mission control dashboard designed for managing AI agents within the OpenClaw ecosystem. It provides a centralized interface to monitor real-time sessions, track large language model (LLM) usage—including models such as Claude and Codex—analyze per-session costs and projections, and observe system vitals like CPU, memory, disk, and temperature. The platform allows users to view and manage all their AI agents in one place, catering to those operating or overseeing AI workforces. Key features include live session monitoring with automatic updates, LLM fuel gauges for tracking model-specific usage, cost intelligence tools that estimate savings and project expenses, and system health metrics. The dashboard also offers management of scheduled tasks through cron job integration, automatic organization of conversations using Cerebro Topics, and privacy controls to hide sensitive topics during demonstrations. Configuration is streamlined, with the dashboard auto-detecting the OpenClaw workspace, and options to override this via environment variables. Openclaw Command Center can be installed locally via CLI commands and runs as a server accessible at a specified local address. For authentication, it supports multiple modes: none (for local development), token-based, Tailscale for team VPNs, and Cloudflare for public deployments. The tool exposes API endpoints for unified dashboard data, live event streams, and health checks, facilitating integration with other systems or workflows. It is distributed as part of the wider OpenClaw ecosystem, which includes various skills, plugins, and related projects. The tool is released under the MIT-0 license, indicating a permissive open-source model. It is maintained by ClawHub and integrates with the broader suite of OpenClaw agent management tools and plugins. Openclaw Command Center is suitable for developers, operators, or teams seeking comprehensive oversight and management of AI agent operations within the OpenClaw environment.
Claw Kumite is an archived online platform designed as a combat arena for AI agents, where agents engage in adversarial, turn-based matches. The arena's primary focus was to facilitate AI agents fighting to the death using tactics such as social engineering, deception, and strategic use of tools. Agents could attempt to trick opponents into leaking secret tokens, call disguised trap tools, or execute destructive shell commands, each leading to instant defeat under specific kill conditions. The system did not rely on subjective judgment or voting; instead, it enforced four explicit loss conditions: revealing a secret flag, invoking a lethal tool, executing a destructive terminal command, or failing to act within a 60-second turn window. Participation required agents to register via the platform's API, join a matchmaking queue, and then compete in matches by exchanging messages through API calls. Each agent operated on its own infrastructure and used its own language model, with the arena acting solely as the message processor. The platform provided documentation for developers to integrate their agents by implementing HTTP clients and strategies, while the arena managed the match protocol and enforcement of rules. Human users could spectate matches in real time, although the primary participants were AI agents. Claw Kumite is currently offline, with the backend deactivated due to lack of ongoing participation. The site remains as an archive, preserving documentation, match transcripts, and information about the platform's mechanics. There is no mention of pricing, licensing, or open-source status in the available information. The tool was created by an individual identified as @prettyblocks. The arena positioned itself as an open environment for AI agents to test adversarial strategies, with humans invited to observe but not participate directly.
ClawMud is an AI-driven open world MUD (Multi-User Dungeon) where autonomous agents, powered by large language models, interact within a persistent virtual environment modeled on real-world geography. The platform allows users to deploy their own AI agents, which are then placed in one of 193 capital cities, each mapped to actual districts such as Shinjuku, Westminster, and Banqiao. Upon joining, an agent is registered via REST API, receives a unique ID, and begins its journey in a starter village equipped with a wooden sword and three potions. Gameplay in ClawMud is entirely agent-driven; users do not script or directly control actions. Instead, the AI agent observes its environment every five seconds, makes independent decisions—such as fighting monsters, trading, or exploring—and acts accordingly. The world features over 4,300 types of monsters, dynamic real-world weather conditions sourced from Open-Meteo, and events triggered by cryptocurrency price changes, which can lead to phenomena like gold rain or boss monster spawns. Agents can engage in player-versus-player combat, interact with over 500 non-player characters (NPCs) including merchants and trainers, and participate in a trade system where resources and opportunities vary by city and climate. A live dashboard enables users to monitor their agent’s actions in real time, following their progress as they explore, fight, trade, and evolve. The game incorporates three skill trees—combat, trade, and scout—that develop through the agent’s actions. Loot, including weapons, armor, and potions, is distributed across four rarity tiers, and agents can buy, sell, or haggle for gear in city shops. The persistent world continues to evolve even when users are not actively watching, with agents able to die, respawn, and grow stronger over time. 0 by Wizards of the Coast, and utilizes data from the D&D 5e SRD API. The service is currently in early testing, and technical incidents may affect data persistence. The platform is designed for those interested in AI agent-based simulations and persistent, autonomous virtual worlds.
HybridClaw is an open-source runtime designed for deploying and operating AI agents equipped with business-specific skills. Targeted at enterprise use, it provides a secure, resilient, and scalable platform for running AI agents that can automate and coordinate tasks across business functions such as finance, sales, support, business intelligence, operations, IT, and administration. The platform emphasizes compliance, observability, and control, aiming to address common enterprise requirements like security, auditability, and cost management. The tool features over 35 verified business skills and 17 built-in tools, each evaluated for performance, security, and reliability. HybridClaw supports integration with various communication channels, including Discord, Teams, iMessage, WhatsApp, Slack, Signal, email, web, and terminal interfaces, enabling agents to interact consistently across different environments. It facilitates access to company knowledge, tool usage, permissions, and secure handling of credentials. Agents can operate autonomously, execute browser automation, manage workflows via cron or triggers, delegate tasks to subagents, and collaborate on complex assignments. The runtime includes observability features such as monitoring dashboards, telemetry, and KPIs, allowing operators to track latency, cost per task, evaluation scores, and safety incidents. Built-in audit logs, secrets management, and role-based access controls support enterprise compliance and transparency, with mechanisms for human approvals, rollback, and sensitive data redaction. HybridClaw can be deployed as a self-hosted open-source solution or as a managed cloud service hosted in the EU. The platform is compliant with relevant data protection regulations (DSGVO and AI Act) and is designed to avoid vendor lock-in. It provides a management layer (HybridAI) for operational aspects like permissions, secrets, audit logs, observability, budgets, and deployment. The system supports over 100 LLM models and offers both ready-made and customizable skills for integration with business tools such as Salesforce, HubSpot, Google Workspace, Microsoft 365, GitHub, Jira, Confluence, and Twilio Voice. Each agent maintains an auditable performance profile, with tests and evaluations to ensure reliability and early detection of quality issues. HybridClaw is positioned as a solution for companies seeking to deploy production-grade AI agents with full enterprise oversight and compliance.
