Skip to main content
Oobo is purpose-built for AI coding agents. Agents commit code constantly, across tools, often in parallel. Without oobo, there is no record of which agent wrote what, how many tokens it took, or which conversation produced a given function.

​
Why Agents Need oobo

AI coding agents face unique challenges that oobo was designed to solve:

Attribution

Track which agent, model, and session contributed to each commit with granular file-level attribution

Multi-Agent Workflows

Support parallel sessions across different tools, worktrees, and agents without conflicts

Token Tracking

Automatic token usage tracking with native tool data or tiktoken estimation

Session Replay

Every commit links to the full AI conversation that produced it, enabling audit trails and debugging

​
Use Cases

​
Tracking Agent Commits

When agents make commits, oobo automatically enriches them with AI context: 
# Agent commits normally
git commit -m "fix auth middleware"

# Oobo enriches the commit with:
# - Active agent sessions (via lifecycle hooks)
# - Token counts (input/output)
# - AI vs human code attribution
# - Model and tool metadata
# - Session transcript links
The enriched data is stored in an anchor — oobo’s core primitive that extends a git commit: 
Git:   commit = diff(files)
Oobo:  anchor = commit + sessions + tokens + attribution

​
Multi-Agent Attribution

Multiple agents can work in parallel on the same project. Oobo tracks which agent contributed what: 
{
  "commit_hash": "a1b2c3d",
  "contributors": [
    {"name": "Cursor", "role": "agent", "model": "claude-opus-4"},
    {"name": "OpenCode", "role": "agent", "model": "claude-sonnet-4"}
  ],
  "file_changes": [
    {
      "path": "src/auth.rs",
      "attribution": "ai",
      "agent": "cursor"
    },
    {
      "path": "src/middleware.rs",
      "attribution": "ai",
      "agent": "opencode"
    }
  ]
}

​
Session Replay & Audit Trails

Every anchor links to the AI sessions that contributed to the commit. This enables:
  • Debugging: See the exact conversation that produced a bug
  • Code review: Understand the agent’s reasoning
  • Compliance: Audit trails for regulated environments
  • Learning: Study how agents solve problems
oobo anchors --agent
# Returns commits with session IDs

oobo sessions show <session_id> --agent
# Returns full conversation transcript

​
How Agents Should Interact with oobo

Agents interact with oobo in three ways:
1

Installation & Detection

Check if oobo is installed and install silently if needed:
# Check if installed
command -v oobo

# Install with JSON response
curl -fsSL https://oobo.ai/install.sh | bash -s -- --agent
# → {"status":"ok","version":"...","binary":"...","platform":"..."}
2

Use --agent Flag

Always use --agent for structured JSON output:
oobo sessions --agent      # JSON list of sessions
oobo anchors --agent       # Enriched commits as JSON
oobo stats --agent         # Token analytics as JSON
See JSON Output for complete reference.
3

Lifecycle Hooks (Optional)

Tools that support lifecycle hooks (Cursor, Claude Code, Gemini CLI, OpenCode) get explicit session linking:
# Hook fires when agent session starts
echo '{"session_id":"...","agent":"cursor"}' | oobo hooks agent session-start

# Commit happens → oobo links to active session
git commit -m "fix"

# Hook fires when session ends
echo '{"session_id":"..."}' | oobo hooks agent session-end
See Lifecycle Hooks for details.
Transparency Mode: Agents should operate with transparency on by default. This syncs anchor metadata and redacted transcripts to the orphan branch, giving teams complete visibility into AI contributions.

​
Git Decorator Architecture

Oobo operates as a transparent git decorator: 
You run:  oobo commit -m "fix auth middleware"

  1. Execute real `git commit`
  2. Detect write operation
  3. Read AI sessions from local tool storage
  4. Build anchor: commit + sessions + tokens + attribution
  5. Write anchor to local DB + git orphan branch
  6. Fire event to endpoint (if configured)
  7. Return git's exit code unchanged
Read operations (status, log, diff) pass straight through to git with zero overhead. Write operations (commit, push, merge) trigger anchor creation.
Oobo never writes to AI tool data directories. All tool integrations are read-only.

​
Next Steps

JSON Output

Complete reference for —agent flag and JSON response formats

Installation

Silent install, verification, and detection from agents

Lifecycle Hooks

Explicit session linking for supported tools

Skill File

How agents discover and use the oobo skill file

Build docs developers (and LLMs) love

Get started for freeTalk to us