Minimo — observability TUI for AI coding agents

Observability TUI for AI coding agents — monitors Claude Code, OpenCode, Codex and Kimi sessions in one view, tracking context, cost and token burn in real time. Ships as a single Go binary.

What I built

Built an open-source Go terminal UI that monitors AI coding agent sessions across Claude Code, OpenCode, Codex and Kimi-code from a single view.
Normalised divergent per-provider session formats into one schema to track context usage, cost and token burn rate in real time.
Shipped as a single self-contained binary with no runtime dependencies, published under an open-source license.

minimo — 8 sessions (active/idle) · 1 active · ~$369.62 API-equiv
claude-code ✓   opencode ✓   codex ✗ (~/.codex not found)   kimi-code ✗

  STATUS  PROVIDER     MODEL              CONTEXT               LIFETIME   COST       RATE
  ●       claude-code  claude-sonnet-4-6  [████████░░] 14%/1M   742.5M     ~$369.62   73K/m
  ○       claude-code  claude-opus-4-8    [████░░░░░░]  8%/1M   40.1M      ~$34.94    31K/m
  ○       opencode     deepseek-v4-flash  -                     3.9M       $0.0000    15K/m

Sample output from the README.

How it works

Reads each agent's own session files directly, so it needs no API keys and no config.
Claude Code: JSONL transcripts under ~/.claude/projects, with exact per-turn token counts.
OpenCode: its local SQLite database, including the exact cost the provider reports.
Codex and Kimi Code: JSONL transcripts under ~/.codex and $KIMI_CODE_HOME.
Costs are estimated from the LiteLLM pricing catalog (fetched live, with an embedded snapshot as fallback); exact provider costs are never overwritten.

Commands

minimo: the live dashboard. Navigate with ↑↓ or j/k, press Enter for session detail, h for history and s for usage stats.
minimo status: a one-shot table; --watch refreshes on file changes and --json pipes into jq, scripts or Grafana.
minimo stats: per-model cost and time utilisation over the last 24 hours, 7 days and 30 days.

Cost, explained

Minimo shows API-equivalent cost: what the tokens would cost at pay-as-you-go rates.
On a flat Claude Pro or Max subscription that figure is the value of your usage, not a bill, and the session detail shows it as a multiple of the plan price.

What sets it apart

Compared with ccusage, abtop and tokentop, it adds a subscription-ROI line, cross-provider cost per 1K tokens, subagent token attribution and --json output.
Ships as a single Go binary with no CGO, or a ~10 MB distroless Docker image with read-only mounts.

Source on GitHub

Built by Nijeesh NJ, AI Engineer in Bengaluru. See more projects.