aglogs (Agent Logs) is a command-line tool for parsing, monitoring, and analyzing transcripts from local AI agents. It normalizes proprietary log formats from multiple providers into a unified structure, enabling consistent inspection and integration with the Grove ecosystem.

Core Mechanisms

Unified Normalization: The tool reads provider-specific log formats (typically JSONL) and converts them into a standard UnifiedEntry structure. This normalization handles differences in timestamp formatting, role definitions, and tool call representations (e.g., merging Claude’s separate tool use and result events).

Discovery Strategies: aglogs locates session transcripts using a tiered approach:

1. Direct Path: Accepts direct file paths to log files.
2. Session Registry: Queries the grove hooks database (~/.local/state/grove/hooks/sessions/) to resolve Grove Job IDs to native agent session IDs.
3. Filesystem Scanning: Scans default storage directories for supported providers to index available sessions.

Output Formatting:

• Human-Readable: Renders formatted text with syntax highlighting, diff views for file edits, and distinct icons for tools vs. conversation.
• JSON: Outputs raw structured data for machine consumption.

Supported Providers

aglogs supports the following local agent runtimes:

• Claude Code: Scans ~/.claude/projects/ for project-scoped JSONL transcripts.
• Codex: Scans ~/.codex/sessions/ for session logs.
• OpenCode: Assembles fragmented message and part files from ~/.local/share/opencode/storage/.

Integration with Grove Flow

aglogs serves as the observability layer for flow orchestrations.

Live Streaming: The flow plan status TUI executes aglogs stream to display real-time output from running agents directly within the terminal interface. This allows monitoring of headless and interactive agents without attaching to their specific processes.

Transcript Archival: Upon job completion, flow executes aglogs read to retrieve the full session history. This content is appended to the job’s Markdown file (e.g., 01-impl.md), preserving the chain of thought and execution history alongside the task definition.

Session Resumption: When resuming an interactive job, flow uses aglogs get-session-info to look up the native agent session ID associated with the job, enabling the agent to continue context from the previous run.

Features

• aglogs list: Indexes and displays available sessions across all providers. Supports filtering by project or job name.
• aglogs read: Outputs the full log for a specific session ID, Job ID, or file path.
• aglogs stream: Tails a live log file, rendering new entries as they are written.
• aglogs query: Filters messages within a transcript by role (e.g., --role user) or content.