AgentHUD

A heads-up display for your AI coding agents — Claude Code, OpenAI Codex, AWS Kiro, and opencode. AgentHUD reads each agent's on-disk sessions (JSONL files, or opencode's SQLite store) and merges them into one tree, so a project you've touched from several agents shows as a single row with combined sessions and sub-agents.

It's organized as three layers, by who's reading — a human at a glance, a human at the end of the day, and a machine:

• agenthud — the live HUD. A real-time TUI showing which session is [working] versus [waiting] on you, right now, across every agent. Glance at it in a side terminal and know what all your agents are doing. This is the hero — the name is HUD for a reason.
• agenthud summary — the daily digest. Fold a day (or a week) into an LLM engineering summary — a standup-in-one-command habit, via whichever agent CLI you have installed (claude, codex, or kiro-cli, auto-detected).
• agenthud report — the machine layer. Structured Markdown / JSON. summary is literally report piped into an agent CLI, so the daily digest isn't a black box — and you can pipe report into anything else, too.
• agenthud follow — the live machine feed. A chronologically-merged stream of every event (activity + state + lifecycle) across all sessions and sub-agents, as human lines or --json NDJSON — the read-only substrate a higher-level supervisor can consume. See FEATURES.md.

→ See FEATURES.md for the full surface — every flag, keybinding, config key, file path, and env var. Per-agent session schemas: Claude Code · Codex CLI · Kiro IDE · Kiro CLI · opencode (or browse docs/schemas/).

Requires Node.js 20+ (opencode sessions need Node 22+ — they're read via the built-in node:sqlite; on older Node, opencode is simply skipped and everything else works). Open agenthud in a separate terminal while you work; press ? inside the TUI for in-app help.

Try without installing

npx agenthud
# or: bunx agenthud

Install for daily use

npm i -g agenthud
# or: bun i -g agenthud

Platform support. Primary development is on macOS and Linux; the full test suite runs on all three platforms in CI (including Windows). Windows runtime behavior is exercised by a manual smoke job but isn't daily-driven — issues there are valued bug reports.

Quickstart

# 1 · The live HUD
agenthud                                  # all projects, every agent
agenthud --cwd                            # scope to the project containing $PWD

# 2 · The daily digest
agenthud summary                          # synthesize today via your agent CLI
agenthud summary --last 7d                # cross-day synthesis of the last 7 days
agenthud summary -oI                      # open the summary + the summaries index

# 3 · The machine layer
agenthud report                           # today's activity as markdown
agenthud report --format json             # script-readable
agenthud report --with-git                # merge git commits into the timeline

The live HUD — agenthud

The hero. A split-view TUI: a project tree (top) and an activity viewer (bottom), refreshing as your agents work.

┌─ Projects ────────────────────────────────────────────────────────────────────────────────────┐
│ 4 projects (2) · 31 sessions (3) · 142 sub-agents (1) · ⊘ 2 hidden                            │
│ > agenthud  ~/WestbrookAI/agenthud  6 sessions (2) · 114 sub-agents                           │
│     #864f [working] Fix the auth bug in login flow                         9s 41% claude opus │
│         ├─ » code-reviewer                                                                    │
│     #019e [waiting] review the data layer                                  2m 44% codex gpt-5 │
│   myproject  ~/work/myproject                                                              2d │
│     #def4 [cool] Add OAuth support                                        2m 5% kiro-ide auto │
│ ... 12 cold projects                                                                          │
└───────────────────────────────────────────────────────────────────────────────────────────────┘
┌─ Activity · agenthud ─────────────────────────────────────────────────────────────────────────┐
│ [10:23] ○ Read  src/ui/App.tsx                                                                │
│ [10:23] ~ Edit  src/ui/App.tsx                                                                │
│ [10:23] $ Bash  npm test                                                                      │
│ [10:23] < Response  Tests passed successfully                                                 │
│ [10:25] ⠧ Edit  src/auth/oauth.ts  ← bold + spinner = live                                    │
└───────────────────────────────────────────────────────────────────────────────────────────────┘

The activity viewer (bottom) follows your selection — pick any session or sub-agent and it streams what that node is doing right now, the live line bold with a spinner. So when a sub-agent is mid-task you can watch exactly what it's reading, editing, or running; press ↵ on any activity for a scrollable detail view.

The badge tells you, per session, what the agent is doing right now:

• [working] — there's a pending tool call at the tail; the agent is mid-step.
• [waiting] — the turn yielded back to you (a question, or a finished reply).

These live states are read from the structure of each session's JSONL tail and override the time-based recency badges ([hot] / [warm] / [cool] / [cold]) whenever a session is live. Each row also carries a colored provider label (claude / codex / kiro / kiro-ide / opencode), its model, and a context-window gauge; only hot/warm (and live) sessions count as active and render bright. Sub-agents nest under their parent regardless of which agent spawned them, the panel title is a tree-wide census, and cold projects collapse under a ... N cold projects sentinel.

Full keybinding and badge reference: FEATURES.md.

The daily digest — agenthud summary

The asynchronous human layer: a day or a date range, synthesized into an engineering summary you can read like a standup note.

agenthud summary --date yesterday         # one day
agenthud summary --last 7d                # a rolling window
agenthud summary --from 2026-06-01 --to 2026-06-07

The summary runs through an agent CLI of your choice. By default it auto-detects the first one installed in claude → codex → kiro order; set a default with summary.engine in ~/.agenthud/config.yaml, or override per-run with --engine <claude|codex|kiro>. So Claude-, Codex-, and Kiro-only users all get summaries with no extra setup.

Summaries are cached one-file-per-day under ~/.agenthud/summaries/, cross-linked into a browsable index (-I to open it), and stamped with the engine + model that produced them — so switching engines regenerates rather than serving stale text.

Scripting & integration — agenthud report

agenthud report is the machine-readable layer the other two are built on. It isn't a feature you rarely touch — it's the substrate of the one you use every day:

# This is, essentially, what `summary` does under the hood:
agenthud report --date yesterday | claude -p "$(cat ~/.agenthud/summary-prompt.md)"

# …so the same report can flow anywhere else:
agenthud report --format json | jq '.sessions[].model'   # into a dashboard
agenthud report --with-git    | your-own-llm-call        # into any pipeline

summary is just report piped into an agent CLI. The digest isn't a black box — it's report + a prompt + whichever LLM you like, and you can swap any part. Markdown is the default; --format json (with provider + model per session) is the script-friendly form.

Configuration

~/.agenthud/config.yaml is auto-created on first run with sensible defaults. CLI flags override config values per-invocation. Resolution order is CLI flag → summary.<key> → report.<key> → built-in default, and the effective values print to stderr at the start of every report / summary run.

App-managed UI state (hidden projects/sessions/sub-agents toggled by h) lives separately in ~/.agenthud/state.yaml.

Full schema, file paths, and env vars: FEATURES.md → Config.

More

• Reference: FEATURES.md — every flag, keybinding, config key, file path, env var
• Feature history: FEATURE-HISTORY.md — headline features by version
• Release history: CHANGELOG.md
• Deferred items: BACKLOG.md
• Issues / PRs: GitHub

License

MIT