Build AI Agents Without Code Using Agent Forge.
Agent Forge is Prediction Guard's web-based platform for building and deploying specialized AI agents complete with knowledge bases, MCP tool integrations, enterprise security controls, and custom configurations. No code required.
Describe Your Agent, Agent Forge Does the Rest
Type what you want your agent to do. Agent Forge generates the system prompt, configures settings, and has it ready to chat in seconds.
Model Selection, System Prompt & Security All in One Form
The agent configuration form covers everything. Choose from available LLMs, write a system prompt that defines the agent's expertise, configure streaming, and set enterprise security controls all without code.
- Available models:
gpt-oss-120b,gemma-3-27b-it,Qwen2.5-Coder-14B,Hermes-3-Llama - System prompt defines role, expertise, responsibilities, knowledge scope, and behavioral guidelines
- Streaming mode for interactive conversations; Non-Streaming required when using MCP tool calls
- Built-in PII detection, injection protection, and topic filtering enabled per agent
Fine-Tune Temperature, Top P, Top K & Max Tokens
Advanced settings let you precisely control how your agent generates responses from strict, deterministic outputs for compliance tasks, to creative responses for brainstorming.
- Temperature 0.0–0.2 for compliance and data extraction, 0.7–1.2 for creative writing and brainstorming.
- Top P Nucleus sampling. Use 0.9–0.95 for most applications, 0.5–0.7 for more focused responses.
- Top K Limits token sampling. Use 10–20 for focused outputs, 40–50 for balanced variety.
- Max Output Tokens 100–300 for quick answers, 500–1000 for standard replies, 1500–2500 for long-form content.
For compliance or safety-critical applications, use lower temperature (0.0–0.3) to ensure consistent, accurate responses.
Connect Agents to Documents, Databases & Live APIs
Give your agents organizational knowledge via RAG document uploads, and live tool access via native MCP integration at both the organization and individual agent level.
- Knowledge Base Upload PDFs, text files, and docs. Automatic embedding, vector search, configurable relevance thresholds.
- Org-Level MCP Admins configure shared MCP servers available to all agents (databases, internal tools, APIs)
- Agent-Level MCP Individual agents connect to specialized MCP servers for unique workflows
- File upload support in chat users share PDFs and documents directly in conversations
Full Visibility Into Every Agent, Thread, and Message.
Enable the protections and topic boundaries that keep every agent aligned with your risk posture: PII detection, injection blocking, and topic filtering configured per agent without code.
- Prompt Injection Protection Configurable sensitivity from strict to permissive, blocking malicious instructions before they reach the model.
- PII Detection & Replacement Automatically detect and replace sensitive personal data with random fake values before it enters the context window.
- Topic Filtering Define exactly what subjects the agent is allowed to discuss, with a custom off-topic prompt for out-of-scope requests.
- Per-Agent Controls Every guardrail is scoped to the individual agent, different risk profiles for different use cases, all from one platform.
See What's Possible. Try a Live Agent.
The Complete Platform for Building Governed AI Agents
RAG Knowledge Base
Build Agents Your Organization Can Trust.
Agent Forge gives every team the tools to build, deploy, and manage compliant AI agents, governance baked in from the first message.