Agent Workspace

Your agent has persistent storage mounted at /data. Files here survive between conversations and container restarts.

Directory structure

/data
├── MEMORY.md           # Agent's persistent memory
├── config.json         # Agent preferences (model, etc.)
├── sessions/
│   └── {id}.jsonl      # Conversation history files
├── outbox/
│   ├── email/          # Pending emails to send
│   ├── sent/           # Successfully sent emails
│   └── failed/         # Failed email attempts
└── events/
    └── {timestamp}.jsonl  # Run events for dashboard

MEMORY.md

This is your agent’s brain. It persists across sessions and contains:

Agent identity and configuration
Owner information
Outbox pattern instructions
Custom context you want the agent to remember

The agent reads this file at the start of every turn.

config.json

Stores agent preferences like model selection:

{
  "model": "haiku"
}

You can change the model via email:

/model haiku — Fastest, cheapest
/model sonnet — Balanced
/model opus — Most capable

Sessions

Each conversation is stored as a JSONL file in /data/sessions/. Sessions contain:

Your messages
Agent responses
Tool calls and results
Usage statistics

Managing sessions

Command	Effect
`/clear`	Start a new session (preserves old ones)
`/resume`	List recent sessions with hashes
`/resume abc1`	Resume session by hash
`/compact`	Compress context to reduce token usage

Outbox

The outbox is how your agent sends emails. Instead of calling APIs directly, the agent writes JSON files:

/data/outbox/
├── email/          # Pending - agent just wrote these
├── sent/           # Successfully sent by platform
└── failed/         # Failed to send (will retry)

Example outbox file:

{
  "to": ["[email protected]"],
  "subject": "Re: Your question",
  "body": "Here's what I found...",
  "in_reply_to": "<original-message-id>",
  "status": "pending"
}

See Outbox Pattern for details.

Events

The /data/events/ directory stores real-time events from agent runs. The dashboard uses these to show:

Tool calls in progress
Streaming text output
Errors and completions

Pre-installed tools

Your agent’s container includes:

Core utilities:

git, curl, wget, jq, yq
ripgrep (rg), fd-find (fdfind), tree
sqlite3, psql (PostgreSQL client)
zip, unzip, tmux, less

Python:

python3 with uv (fast package manager)
fpdf2 for PDF generation

Developer tools:

gh (GitHub CLI)
Node.js 22 LTS

Skills

Skills are documentation files that teach your agent how to use specific tools. They’re cloned to /opt/skills/ in the container:

/opt/skills/pi-skills/ — Core skills from pi-mono
/opt/skills/tiny-skills/ — TinyFat platform skills

See Skills & Tools for the full list.

Storage limits

Your agent’s R2 storage is currently unlimited during launch. In the future:

Free tier: TBD
Tiny tier ($5/mo): TBD

Files in the outbox are automatically cleaned up after successful sending.

Accessing files

You can view and edit files through:

Email — Ask your agent to read/write files
Dashboard — File browser in the web UI (coming soon)
Direct R2 — Export your data anytime

Next steps

Memory Bank Pattern — Organize project knowledge
Skills & Tools — What your agent can do
Email Communication — How email works