Follow the conversation alongside tool calls and file activity. Search recorded messages and open the matching event.

TMA1 / AGENT OBSERVATORY
Sessions, tool calls, and token usage in one local workspace. Inspect what happened, compare runs, and feed observations back into the agent.
Claude Code · Codex · Copilot CLI · OpenCode
Follow the conversation alongside tool calls and file activity. Search recorded messages and open the matching event.

Compare two sessions across agents: reported tokens, cache read share, tool failures, and compaction events. Missing usage stays unavailable.

Overview summarizes recorded tool completions and failures across agents. Claude Code adds tool latency, success rates, and failure details from its telemetry.

Inspect token usage, cache activity, and model costs where telemetry provides them. Estimates depend on available usage and pricing; they are not billing totals.

Review repeated build or test failures, stale file views, external changes, and large input-usage samples. Findings provide evidence and a suggested next step.

Inspect recorded commands, URL requests, and pattern-based risk flags in GenAI traces. Flags are review signals, not a security verdict.

Screenshots and video use demo data.
Claude Code and Codex hooks provide session context and advisory findings. Prompt review can use a configured model to suggest changes backed by conversation excerpts.
MCP tools and the /tma1-peer and /tma1-search skills retrieve earlier conversations, including work from other agents on the same project.
Paste this instruction into Claude Code or Codex
Read https://tma1.ai/SKILL.md and follow the instructions to install or upgrade TMA1 for your AI agent
# macOS / Linux
curl -fsSL https://tma1.ai/install.sh | bash
# Windows (PowerShell)
irm https://tma1.ai/install.ps1 | iexOpenTelemetry SDKs: send OTLP/HTTP to http://localhost:14318/v1/otlp. GenAI traces use the OpenTelemetry GenAI semantic conventions.
OpenCode imports sessions through a plugin or an exported snapshot. Import does not install hooks or inject context. Copilot CLI session logs are discovered automatically.
OpenCode setup →Traces and conversations are stored in ~/.tma1/data/, backed by GreptimeDB. Optional model reviews send selected prompts and conversation excerpts to the configured provider.
Read the integration docs →