TMA1 is a local-first observability tool designed to operate directly within the feedback loop of AI agents. It records each large language model (LLM) call locally and routes observations back into the agent’s next turn, enabling agents to act on specific feedback in real time rather than relying solely on dashboards for later review. The tool is intended for scenarios where agents interact with code, files, or external tools, and where immediate, actionable insights can improve agent performance and reliability.
TMA1 delivers a range of features focused on closed-loop perception and cross-agent collaboration. It monitors for repeated failures, stale file views, and broken builds, and when certain rules are triggered, it writes concrete suggestions or fix paths into the agent’s next prompt. The tool supports hooks, anomaly detection, and can block further actions on high-severity issues to prevent silent errors. Peer session capabilities allow agents, such as Claude Code and Codex, to read each other's feedback and reviews on the same project files, facilitating knowledge sharing without manual copy-paste. Anomalies are flagged with direct links to relevant sessions, and a session overlay provides detailed breakdowns of file activity, context, API calls, and a full event timeline. Additional analytics include tool latency, call counts, success rates, cost breakdowns by model, and cache hit ratios. Security monitoring flags shell commands, URL fetches, and injected prompts, and for specific agents like OpenClaw, it tracks webhook errors and stuck sessions. Full-text search across sessions is also supported.
The tool is delivered as a single Go binary and does not require Docker, cloud infrastructure, or external system packages. tma1/, and no data is uploaded to remote services. The dashboard can be accessed at localhost:14318, and the tool works with agents such as Claude Code, Codex, OpenClaw, or any OpenTelemetry (OTel) SDK. GitHub Copilot CLI sessions are auto-discovered. Installation is handled via a single command for macOS, Linux, or Windows, and configuration involves pointing the agent to a local OTel endpoint. TMA1 is designed for developers and teams working with AI agents who require local, actionable observability without sacrificing privacy or control over their data.
TMA1 sits in PulseGate's LLM eval & observability category. It focuses on enabling AI agents to observe and learn from their own actions and failures locally without cloud dependencies. It is built as an open-source project for AI agent developers. TMA1 is open source under the Apache-2.0 license. It runs on the web and the command line, and it can be self-hosted.
TMA1 first shipped in 2026. Development happens publicly on GitHub with 96 stars. Key capabilities include local observability, agent feedback loop, and anomaly detection.
Latest indexed changes and source events
Other apps tracked under the same category.
