Agent CLI

The Agent CLI (fastfold) is an autonomous research agent for drug discovery. Ask questions in natural language, and it plans the analysis, selects the right tools, executes them, validates results, and returns data-backed conclusions. It integrates with Fastfold AI Cloud for GPU compute, protein folding, and workflow orchestration.

• PyPI: fastfold-agent-cli
• GitHub: fastfold-ai/fastfold-agent-cli

Installation

Requires Python 3.10 or later.

Install uv (recommended)

If you already have uv installed, skip to the next step.

curl -LsSf https://astral.sh/uv/install.sh | sh

Official uv installation docs: Astral uv installation guide

uv tool install "fastfold-agent-cli[all]" --python 3.10

Upgrade

To upgrade an existing Agent CLI install with uv:

uv tool install "fastfold-agent-cli[all]" --python 3.10 --upgrade

Authentication

fastfold setup

Quick start

# Start interactive session
fastfold

# Single query
fastfold "What are the top degradation targets for this compound?"

# Validate your setup
fastfold doctor

Example queries

Target prioritization

fastfold "I have a CRBN molecular glue. Proteomics shows it degrades IKZF1, GSPT1,
          and CK1α. Which target should I prioritize?"

Protein folding

fastfold "Fold this sequence with boltz-2 and find the binding pockets: MALWMRLLPLL..."

Combination strategy

fastfold "My lead compound is immune-cold. What combination strategy should I use?"

Interactive mode commands

Inside fastfold interactive mode:

During an active generation:

• /interrupt requests a graceful stop
• Ctrl+C once does the same interrupt request
• Ctrl+C again quickly (or /interrupt!) force-stops the current run

Background task workflow

Use background tasks for long-running waits and keep working in the same session.

Can you run the BoltzGen `vanilla_target_binding_site` example with one target and one design budget, then keep the wait step in the background so we can continue? Please share the `workflow_id` and output file path.

Then monitor watcher state from the prompt:

/tasks
/tasks refresh

If you're using Warp Terminal, completed background tasks can also trigger desktop notifications (when Warp notifications are enabled in your OS and Warp settings). See Warp Desktop Notifications.

You can tune how long the watcher keeps listening for background tasks:

fastfold config set agent.background_watch_timeout_s 10800

Recommended pattern for job/workflow skills (fold, BoltzGen, MD):

• run submit/create steps in foreground and return job_id or workflow_id immediately
• run only long wait or log-watch steps in background
• use /tasks to verify watcher status while you continue other queries

Tool categories

fastfold tool list    # see all 190+ tools
fastfold skill list   # see loaded skills

Data management

fastfold data pull depmap    # DepMap CRISPR, mutations, expression
fastfold data pull prism     # PRISM cell viability
fastfold data pull msigdb    # Gene sets

# Or point to existing data
fastfold config set data.depmap /path/to/depmap/

Reports

fastfold report list      # list saved reports
fastfold report publish   # convert latest .md to .html
fastfold report show      # open in browser

Share to Slack from interactive mode

If you connected Slack in Fastfold Cloud and configured the agent_cli_report channel, you can share the current session report directly:

/export-share

What it does:

• exports your current session to markdown
• sends it to your configured Slack report channel
• saves a copy to your library
• adds a "View in Library" link in Slack when available

Troubleshooting

Next steps

• SDK for programmatic workflows → SDK Install
• Skill-based agent integrations → Skills
• PyMOL Agent integration → PyMOL Agent