fred-runtime 2.0.4

Runtime adapters and infrastructure wiring for Fred v2 agents.

fred-runtime

fred-runtime is the infrastructure and execution layer for Fred agent pods.
It wires platform services (databases, LLM routing, MCP, observability, security) into the execution engine defined by fred-sdk, and provides a ready-to-use FastAPI app factory so any agent pod is just a registry and a YAML file away from running.

---

Where fred-runtime fits

Fred follows a strict three-layer model:

fred-core          Pure utilities — model factories, embeddings, logging, KPI store
     │
fred-sdk           Execution engine + authoring surface (pip-installable, no infra)
     │              ReAct runtime, Graph runtime, agent contracts, tool abstractions
     │
fred-runtime       Platform adapters + pod factory (this package)
                    SQL checkpointer, MCP wiring, LLM routing, OpenAI-compat surface,
                    FastAPI app factory, observability, security middleware

Rule of thumb:

• Write agent logic in fred-sdk.
• Write infrastructure adapters (DB, MCP server, Keycloak, object store) in fred-runtime.
• fred-sdk must stay importable on a bare laptop with no services running.

---

What's in the box

fred_runtime.app — Agent pod factory

The main entry point for building a Fred agent pod.

from fred_runtime.app import create_agent_app, load_agent_pod_config

config = load_agent_pod_config()          # reads ENV_FILE + CONFIG_FILE (configuration.yaml)
app    = create_agent_app(registry=REGISTRY, config=config)

create_agent_app returns a FastAPI application that exposes:

Method	Path	Description
POST	{base_url}/agents/execute	Single-turn execution — returns final JSON
POST	{base_url}/agents/execute/stream	Streaming SSE execution — yields RuntimeEvent objects
GET	{base_url}/agents	List registered agent IDs
GET	{base_url}/agents/sessions	List session IDs for a user
GET	{base_url}/agents/sessions/{id}/messages	Full conversation history for a session
GET	/v1/models	OpenAI model list (agent IDs as model names)
POST	/v1/chat/completions	OpenAI chat completions — works with Open WebUI, openai-python SDK, etc.

The OpenAI-compatible /v1 surface is enabled by default.
Set app.openai_compat: false in configuration.yaml to disable it for internal pods.

Multi-turn continuity and HITL (human-in-the-loop) are handled transparently through the SQL checkpointer. The session ID is the LangGraph thread_id.

---

fred_runtime.runtime_support — Infrastructure adapters

Module	What it provides
sql_checkpointer	Durable LangGraph checkpointer backed by SQLite (dev) or PostgreSQL (prod)
user_token_refresher	Transparent Keycloak token refresh for long-lived agent sessions
request_context_helpers	FastAPI dependency helpers for extracting user/session context

---

fred_runtime.model_routing — LLM routing

Multi-provider model resolution with per-agent tuning overrides.
Supports routing by agent ID, execution category, or explicit model name.
Providers: OpenAI, Azure OpenAI, Mistral, Ollama, and any LangChain-compatible backend.

---

fred_runtime.common — Knowledge Flow + MCP clients

HTTP clients that connect agent tools to the Fred platform services:

Client	Connects to
kf_http_client	Knowledge Flow REST API (generic)
kf_vectorsearch_client	Vector search / retrieval
kf_markdown_media_client	Document content (Markdown + media)
kf_workspace_client	Workspace and library management
kf_logs_client	Audit log retrieval
kf_fast_text_client	FastText classification
mcp_runtime / mcp_toolkit	MCP server lifecycle and tool injection
context_aware_tool	Tool base class that propagates the runtime context (user, team, token)

---

fred_runtime.integrations — v2 runtime adapters

Small adapters that bridge the platform-agnostic fred-sdk contracts to real services available at runtime (chat model factory, checkpointer wiring, MCP discovery).

---

fred_runtime.client — Developer CLI (fred-agents-cli)

An interactive REPL and one-shot client for any Fred agent pod:

# Interactive mode — connects to http://127.0.0.1:8000/api/v1 by default
fred-agents-cli

# Set the current team scope inside the REPL
/team my-team

# Inspect runtime KPIs without Grafana
/kpi
/kpi tool_name=search

# One-shot
fred-agents-cli --agent my-agent "What is the status of cluster A?"

# Keycloak browser login
fred-agents-cli --login

# Start already scoped to one team
fred-agents-cli --team-id my-team

# Override the metrics endpoint used by /kpi
fred-agents-cli --metrics-url http://127.0.0.1:9115/metrics

The target pod URL is resolved from configuration.yaml automatically, or overridden with --base-url / FRED_AGENT_POD_URL.

---

Configuration

Every Fred pod uses the same two-file convention:

File	Purpose
.env (path from ENV_FILE)	Secrets: API keys, DB URLs, Keycloak credentials
configuration.yaml (path from CONFIG_FILE)	App settings: port, base URL, LLM routing, observability, security

Minimal configuration.yaml for a local pod:

app:
  name: "My Agent Pod"
  base_url: "/myapp/v1"
  host: "0.0.0.0"
  port: 8010
  log_level: "info"
  metrics_address: "127.0.0.1"
  metrics_port: 9115
  kpi_process_metrics_interval_sec: 10

observability:
  tracer: logging
  metrics: prometheus

ai:
  knowledge_flow_url: "http://localhost:8111/knowledge-flow/v1"

Full schema: fred_runtime.app.config.AgentPodConfig.

When observability.metrics: prometheus is enabled, create_agent_app(...) starts a dedicated Prometheus exporter on app.metrics_address:app.metrics_port and restores the shared Fred KPI pipeline, including process and SQL pool KPIs.

---

Installation

# Core library (runtime adapters, CLI, model routing)
pip install fred-runtime

# With FastAPI pod factory
pip install fred-runtime[app]

Requires Python 3.12.

---

Building an agent pod

A minimal pod is three files:

main.py

from fred_runtime.app import create_agent_app, load_agent_pod_config
from myapp.registry import REGISTRY

config = load_agent_pod_config()
app = create_agent_app(registry=REGISTRY, config=config)

__main__.py

import uvicorn
from fred_runtime.app import load_agent_pod_config

def main():
    config = load_agent_pod_config()
    uvicorn.run("myapp.main:app", host=config.app.host, port=config.app.port, reload=True)

if __name__ == "__main__":
    main()

registry.py

from fred_sdk.contracts.models import ReActAgentDefinition

class MyAgent(ReActAgentDefinition):
    agent_id = "my-agent"
    ...

REGISTRY = {MyAgent.agent_id: MyAgent()}

See fred-samples for a working reference pod.

---

Related packages

Package	PyPI	Role
fred-core	pypi	Pure utilities — logging, model factories, embeddings, portable observability
fred-sdk	pypi	Agent authoring — ReAct, Graph, tool contracts
fred-runtime	pypi	This package

---

License

Apache 2.0 — see LICENSE.