Topic lineage and actions in the terminal UI

brew install --cask factorhouse/tap/fh

fh skills install --agent claude     # Claude Code (~/.claude/skills)
fh skills install                    # Codex, GitHub Copilot (~/.agents/skills)
fh skills install --agent project    # ./.agents/skills, committed with a repo

fh 1.1

The terminal UI in 1.1 goes from watching a cluster to working on it. Open a topic to see what feeds it and what reads it, then act on connectors, consumer groups and Flink jobs from the same screen. Every change goes through Kpow, with the same permissions, audit log and approval as the web UI.

The seven agent skills ship inside the same binary. Install them with fh skills install and a coding agent such as Claude Code, Codex or GitHub Copilot can answer operational questions through fh: which topics are biggest or idle, whether a consumer group is healthy, which groups are really lagging, why a service is not receiving data, whether Connect is healthy, where a topic’s data comes from and goes, and how to replay a topic safely. They are unchanged in 1.1.

Requirements

• Kpow 96.5 or later. The TUI connects to a socket that 96.5 adds to the API server.
• The Kpow API enabled. It is off by default. Set API_ENABLED="true"; it serves on port 3001 (API_PORT). See the Kpow API reference for configuration and API authentication.

Topic lineage

enter on a topic lays out its lineage left to right: the source connectors that write to it, the topic, then the sink connectors, consumer groups and Kafka Streams apps that read it. Each node is coloured by health, so a failed connector task shows red even when the connector reports RUNNING. The panel across the top shows the selected node: for the topic, its size, write rate over the last hour, schema subjects and naming strategy, and non-default configs; for a consumer group, its lag by partition; for a connector, its tasks.

Kafka Connect and Flink actions

Restart a connector, just its failed tasks, or one task; pause, resume, stop or delete it; and read the stack trace of any failed task. For Flink jobs, stop with a savepoint, cancel, trigger a checkpoint or savepoint, and read the jobmanager and taskmanager logs, thread dumps and checkpoint history.

Sorting everywhere

Every table sorts with ⇧ and the letter marked in its column header; the same letter again flips the order. That includes the broker list, ACLs, schema subjects and Inspect results, where value fields get their own letters.

Feedback is welcome at [email protected].