FastAIAgent

The Operational Loop

One system. Seven steps.
Zero context switching.

Most teams stitch together 3–4 tools to go from a failing agent to a fixed one. FastAIAgent is the only platform where trace, feedback, eval, optimization, comparison, and deployment happen in a single closed-loop system — so every production failure becomes a shipped improvement, not a fire drill.

⌨ Develop

→

📍 Trace

→

👎 Feedback

→

🧪 Eval Case

→

🤖 Optimize

→

⟷ Compare

→

🚀 Deploy

↩

WITHOUT FASTAIAGENT

4 Hours Per Failure

Failing trace → manually write test case → manually rewrite prompt → manually test → manually deploy. Every production failure is a multi-hour context-switching fire drill across 3–4 different tools.

~4 hrs

WITH FASTAIAGENT

10 Minutes Per Failure

Failing trace → thumbs down → eval case auto-created → Prompt Optimizer rewrites → Compare Mode A/B tests → deploy → SDK pulls new version. All within one system. Zero context switching.

~10 min

Two Entry Points, One Data Plane

Built for engineers.
Designed for teams.

Whether you write Python or build in a UI, you get the same trace dashboard, the same prompt registry, the same eval framework.

⌨ For Engineers

Open-Source SDK

Build agents in Python. Run locally. Connect to the platform for observability, prompt management, and eval — when you're ready.

Multi-provider LLM (OpenAI, Anthropic, Ollama, Azure, Bedrock)
Structured output, streaming, tool calling
Supervisor-Worker multi-agent teams
Chain orchestration with cycles, checkpoints, HITL
Eval framework with 7+ scorers and LLM judge
Local KB (ingestion, chunking, embedding, search)
fa.connect() for platform services

pip install fastaiagent

🖥 For Teams

Full Platform

Build agents in the UI. Visual chain editor. Configure connectors. Test in the Playground. Monitor in the dashboard. No SDK required.

Visual Workflow Editor with drag-and-drop canvas
29+ pre-built connectors and hosted MCP servers
Advanced RAG with hybrid search and reranking
Prompt Registry with auto-optimization
Agent Replay with fork-and-rerun debugging
RBAC, SSO, audit trails, SaaS billing
Air-gapped deployment for regulated environments

Start Free →

Open-Source SDK

Five Lines to Your
First Agent.

The fastaiagent SDK is open-source (Apache 2.0) and runs anywhere Python runs. Start standalone — add the platform later with a single fa.connect() call.

✓Multi-Provider LLM — OpenAI, Anthropic, Ollama, Azure, Bedrock, and custom endpoints
✓Streaming + Structured Output — SSE parsing across providers with response_format support
✓Context/DI — RunContext[T] passes DB connections, API clients, and runtime state to tools
✓Guardrails — 5 types, 5 built-ins. Input, output, tool-call, and tool-result positions
✓fa.connect() — Traces flow up, prompts flow down, evals flow both ways. Platform is optional.

quickstart.py

import fastaiagent as fa

# Connect to platform (optional)
fa.connect(api_key="fa_k_...", project="my-project")

# Define your agent
agent = fa.Agent(
    name="support-bot",
    model="gpt-4o",
    system_prompt="You are a helpful support agent.",
    tools=[search_kb, create_ticket],
    guardrails=[pii_filter, toxicity_check],
)

# Run it
result = await agent.run("I can't log in to my account")

# Traces auto-export to platform
# Prompts pull from registry
# Eval results publish back

Apache 2.0

PyPI: fastaiagent

GitHub

🔀

Supervisor-WorkerSDK

Multi-agent orchestration with context passthrough, streaming delegation, and callable dynamic prompts.

⛓

Chain OrchestrationSDK

Cyclic graphs, checkpointing, HITL gates, parallel execution, and conditional routing — all in code.

🧪

Eval FrameworkSDK

7+ scorers, LLM-as-Judge, trajectory eval, session eval. Publish results to platform with one call.

📚

Local Knowledge BaseSDK

Document ingestion, multi-strategy chunking, embedding, and vector search — fully local, no platform needed.

🔄

Agent ReplaySDK

Fork any execution at any step and re-run with modified inputs. The debugging feature no other framework has.

📋

Prompt RegistrySDK

Version, compose with fragments, pull from platform. PromptRegistry(source="platform") for team-managed prompts.

🔭

OTel TracingSDK

OpenTelemetry spans with BatchSpanProcessor. Local SQLite storage or export to platform via fa.connect().

🛡

5 Guardrail TypesSDK

Input, output, tool-call, tool-result, and content guardrails. 5 built-in guards plus custom definitions.

The Complete Agent Operating System

Everything to build, ship, and improve AI agents.

Everything SDK agents need to go to production — and everything teams need to manage them at scale. One platform. No tool-stitching. No context switching.

ONLY IN FASTAIAGENT

Closed-Loop Operations

Develop → Trace → Feedback → Eval Case → Optimize → Compare → Deploy — in one system. Negative feedback auto-creates eval cases. Prompt Optimizer rewrites. Compare Mode A/B tests. Ship without switching tools.

ONLY IN FASTAIAGENT

Agent Replay

Step through any execution span-by-span. Fork from any point and re-run with modified inputs. Compare original vs. forked outcomes side by side. Time-travel debugging for AI agents.

ONLY IN FASTAIAGENT

SDK + Platform, One Data Plane

Open-source SDK and full platform share the same trace dashboard, prompt registry, and eval framework. fa.connect() bridges them with a single call.

Platform Capabilities

⟐

Visual Workflow Editor

Drag-and-drop canvas with cyclic graphs, conditional routing, parallel execution, HITL nodes, and checkpointing — no orchestration code needed.

📚

Advanced RAG Pipeline

Hybrid search, reranking, query rewriting, adaptive routing, self-grading, contextual enrichment — 20+ configurable retrieval feature flags.

⑂

Prompt Registry & Optimizer

Versioning, fragments, 3-mode auto-optimization, Compare Mode A/B testing, version analytics, and approval workflows.

📍

Observability & Tracing

Full hierarchical traces: LLM calls, tool invocations, tokens, latency, cost — linked to the exact prompt version used.

⚖️

Evaluation Intelligence

LLM-as-Judge scoring, RAG eval (Ragas), annotation queues, online eval policies, A/B comparisons, CI/CD eval API, and MLflow integration.

🧠

Agent Memory

Long-term memory across conversations with automatic extraction. Provider-agnostic — works with any LLM. Configurable thresholds and vector-based recall.

🧬

Synthetic Data Generation

LLM-powered test data from schemas. Preview with any connector, export to eval datasets. Bootstrap testing without manual data collection.

💬

Chat & Conversation Simulator

Streaming playground with tool call visualization. Multi-turn conversation simulator with persona-driven users, session scorers, and trajectory analysis.

🔌

29+ Connectors & MCP

Databases, storage, messaging, CRM, email, and data processing. First-class MCP support with hosted servers.

🛡

Human-in-the-Loop

Approval policies on tool calls. Chain pause for human input. Signed webhooks. Context-aware intelligent rejection.

🔗

Public API & Streaming

RESTful API at /public/v1/. Scoped keys, rate limiting, SSE streaming, feedback, and webhooks.

🔐

Enterprise Infrastructure

SSO + 4-role RBAC, air-gapped deployment, audit trails, encrypted secrets, and EU AI Act compliance via FastAIShield.

29+

Connectors

MCP

Protocol Native

Full API

Embed Anywhere

Air-Gap

Ready to Deploy

Agent Replay — The Feature Nobody Else Has

Time-Travel Debugging for AI Agents.

Step through any agent execution span-by-span. See exactly what the agent thought, which tools it called, what context it had — and fork from any point to explore "what if?" scenarios. Available in both the SDK and the platform.

▶ Replay: Customer Support Agent

Steps: 7 Tokens: 1,847 Cost: $0.0062 Duration: 2.4s

System Prompt

setup · 0ms

LLM Call #1

gpt-4o · 520ms · 412 tok

KB: Search Docs

3 chunks · 85ms

Tool: get_customer

salesforce · 180ms

⑂

Memory Retrieve

2 entries · 42ms

LLM Call #2

gpt-4o · 680ms · 891 tok

Final Response

output · 544 tok

Step 4: get_customer TOOL

Input

{ "tool_name": "salesforce.get_account", "arguments": { "account_name": "Acme Corp" } }

Output

{ "id": "001Dn00000X8kPQ", "name": "Acme Corp", "employees": 520, "annual_revenue": 48000000 }

tokens

1,073 / 1,847

cost

$0.0028 / $0.0062

Step 4 of 7 · ⑂ 1 fork

Enterprise Ready

Built for Regulated Environments.

4-role RBAC, SSO, encrypted secrets, audit trails, air-gapped deployment, and FastAIShield for EU AI Act compliance. No cloud dependency. No vendor lock-in. Your data stays yours.

🔐

SSO + RBAC

Enterprise single sign-on with 4-role permission hierarchy: App Admin, Domain Admin, Developer, Viewer.

📦

Air-Gapped Deploy

Standalone deployment bundles with pre-built images. Zero internet dependency for classified and regulated networks.

📋

Audit Logging

Complete audit trail: who did what, when. Action tracking, resource IDs, actor identification, and IP capture.

🛡

FastAIShieldENTERPRISE

EU AI Act compliance platform. Risk assessment, accountability framework (19 roles), audit trails, and regulatory reporting.

SSO

Enterprise Auth

4-Role

RBAC System

Air-Gap

Ready to Deploy

Zero

Cloud Dependencies

Integrations & Ecosystem

Connects to your stack.

Multi-provider LLM support, 29+ connectors, MCP protocol native, and growing vector DB and observability integrations. No vendor lock-in — bring your own models, tools, and infrastructure.

🧠

LLM Providers

OpenAI, Anthropic, Azure OpenAI, Ollama, AWS Bedrock, Google Vertex, and custom endpoints via enterprise gateway auth.

Multi-Provider

🔌

29+ Connectors

Databases, cloud storage, messaging, CRM, email, data processing, and REST/GraphQL. Each connector is reusable across all agents.

Pre-Built

⚡

MCP Protocol

First-class support for Anthropic's Model Context Protocol. Connect external MCP servers, host your own, and discover tools dynamically.

Native Support

🗄

Vector Databases

Built-in vector store with plans for Qdrant, ChromaDB, Pinecone, Weaviate, and pgvector adapters.

Expanding

📊

MLflow Integration

Export traces and eval runs to MLflow. Bridge AI agent quality with your existing ML experiment tracking infrastructure.

Built-In

🔭

OpenTelemetry

OTel-native tracing with BatchSpanProcessor. Export to any OTel-compatible backend — Datadog, Grafana, or your own collector.

OTel Native

Part of the FastAIFoundry Ecosystem

Agent is just the Beginning.

FastAIAgent is the execution engine within the broader FastAIFoundry product family. Build agents here. Consume them in Workspace. Ground them in context. Govern them at scale.

Build, test, deploy, and run intelligent agents with visual workflows, advanced RAG, prompt optimization, and full lifecycle management.

You are here

FastAIWorkspace

Thin-client consumption portal for business users. Run agents, view structured results, and manage work — no technical skills required.

Live

FastAIContext

The Business Context layer. Knowledge graphs, entity resolution, and semantic understanding that makes agents organisation-aware.

Coming Soon

FastAIShield

AI Governance and Compliance. Risk assessment, audit trails, and regulatory readiness for EU AI Act and beyond.

Live