English Documentation

AnnaAgent provides a reusable core toolkit for the paper AnnaAgent: Dynamic Evolution Agent System with Multi-Session Memory for Realistic Seeker Simulation. The current repository focuses on the core agent runtime, command line interface, local workspace format, memory indexing, model configuration, and optional vLLM-based SFT deployment.

What You Can Do

• Install anna as a terminal command from PyPI.

• Create an isolated AnnaAgent workspace with sample cases and configuration.

• Run full seeker initialization with emotion, complaint-chain, scale, and memory modules.

• Save and reload initialized prompt states for later chat sessions.

• Use base OpenAI-compatible models or local SFT endpoints.

• Download released assets and deploy local SFT models with vLLM on GPU machines.

• Index previous sessions into LanceDB and retrieve long-term memory.

• Build batch experiment states and transcripts for reproducible evaluation.

Recommended Reading Order

1. Quickstart for the shortest working path.

2. Configuration to understand workspaces and secrets.

3. CLI Reference for the main command groups.

4. Local SFT and vLLM Deployment if you need local SFT/vLLM services.

5. Case Data and Memory for case files and LanceDB memory.

6. Troubleshooting when a command fails.

Ethical Scope

AnnaAgent is intended for research and simulation. Do not present generated outputs as clinical advice, do not use the simulator to replace professional care, and do not attempt to reconstruct restricted real counseling records from released artifacts.

Contents

• Quickstart
  • Prerequisites
  • Install the CLI
  • Create a Workspace
  • Configure the Backbone Model
  • Select the Lightweight Model Path
  • Run a Demo Conversation
  • Create a Reusable Prompt State
  • Run Batch Experiments
• Configuration
  • Workspace Files
  • Inspect and Edit Configuration
  • Model Modes
  • Secrets and Safety
• CLI Reference
  • Global Commands
  • Assets
  • Models
  • Data
  • Memory
  • Initialization
  • Runs and Logs
• Local SFT and vLLM Deployment
  • When to Use This Path
  • Create the Deploy Environment
  • Download Released Assets
  • Deploy Services
  • What the CLI Checks
  • Custom CUDA or vLLM Paths
  • Failure Behavior
• Case Data and Memory
  • Case File Shape
  • Create a Sample
  • Long-Term Memory
  • Reusable Prompt State
• Agent Skills
  • Skill Location
  • Public URLs
  • ClawHub-Style Publishing Checklist
  • Attribution
• Project Publishing and Documentation
  • Version Source
  • Build Locally
  • PyPI Publishing
  • GitHub Pages
  • Local Documentation Preview
• Troubleshooting
  • Command Not Found
  • Configuration Fails
  • Model Calls Fail
  • Memory Search Fails
  • vLLM Does Not Start
  • Stale Logs Are Confusing
  • PyPI Install Fails