Skip to content
Alquimia Alquimia Alquimia

Introduction

Alquimia is an enterprise AI orchestration platform that runs large language models and autonomous agents in production. It solves the gap between AI proof-of-concepts and production deployments by providing explicit agent contracts, deterministic execution paths, structured outputs, and enterprise-grade integrations.

The production AI gap

Section titled “The production AI gap”

Most organizations can build an AI demo in days. Getting that demo to production — reliably, securely, and at scale — takes months, if it happens at all. The blockers are predictable:

| Challenge | What breaks in production | |---|---| | Hallucinations | No validation layer between LLM output and downstream systems | | No auditability | Can’t explain what the agent did or why — compliance fails | | Security gaps | API keys in environment variables, no secret rotation, no auth | | Integration complexity | Enterprise systems (ERPs, ticketing, AD) require session-aware, authenticated connectors | | No memory | Every conversation starts from scratch — no context continuity | | Uncontrolled autonomy | Agents execute actions without human oversight or approval gates |

Alquimia addresses every one of these with first-class platform features, not workarounds.

The two components

Section titled “The two components”

| Component | What it is | When you use it | |---|---|---| | alquimia-core | Python SDK — the agent execution engine | Embed agent logic in any Python application | | alquimia-runtime | FastAPI service — HTTP API wrapping the core | Deploy agents as a standalone service consumed over HTTP |

In a typical enterprise deployment, alquimia-runtime runs as a Kubernetes workload and uses alquimia-core internally. Your applications interact with it over HTTP using bearer tokens or Keycloak OIDC.

Core principles

Section titled “Core principles”

Explicit over implicit. Every agent has a declared system prompt, tool list, memory strategy, and evaluation strategy. Nothing is inferred at runtime. Agent configurations are versioned JSON documents stored in an OCI-distributable registry.

Determinism over magic. The execution loop is a typed state machine. Every state transition is a Pydantic model. You can inspect, log, and replay every step. The worklog is an append-only audit trail of every action the agent took.

Observability over hope. OpenTelemetry metrics, structured logging, SSE streaming, and X-Request-ID propagation give you full visibility into what the agent is doing and why — at every layer of the stack.

Enterprise constraints are features. Vault secrets, Keycloak auth, OCI artifact distribution, human-in-the-loop approval gates, and Kubernetes-native deployment are first-class concerns, not afterthoughts.

What Alquimia is not

Section titled “What Alquimia is not”
  • Not a chatbot framework. It is an execution layer for agents that do real work — query databases, call APIs, delegate to other agents, and produce structured outputs.
  • Not a LangChain replacement. LangChain is used selectively for LLM connectors and vector stores. Alquimia owns orchestration, memory, and tool governance.
  • Not a no-code tool. It is a developer platform with a Python SDK and a REST API. Configuration is code — versioned, validated, and distributed.
  • Not a managed service. Alquimia runs in your infrastructure, under your security controls, with your data never leaving your environment.

Enterprise use cases

Section titled “Enterprise use cases”

Customer support automation — Agents with access to CRM, ticketing, and knowledge bases. Human-in-the-loop approval for sensitive actions. Full conversation audit trail.

Internal IT operations — Agents integrated with IBM Maximo, Active Directory, and internal APIs. Session-aware connectors with Vault-backed credentials. Multi-channel input (Slack, Email, WhatsApp).

Document intelligence — RAG pipelines with Qdrant vector stores. Chain-of-Density summarization for long documents. Structured output extraction with JSON schema validation.

Multi-agent workflows — Orchestrator agents that delegate to specialist agents via A2A (Agent-to-Agent) protocol. Depth-limited recursion with configurable governance.

Competitive positioning

Section titled “Competitive positioning”

| Capability | Alquimia | LangChain/LangGraph | AutoGen | Custom build | |---|---|---|---|---| | Enterprise auth (Keycloak, JWT) | ✅ Built-in | ❌ DIY | ❌ DIY | ❌ DIY | | Vault secret management | ✅ Built-in | ❌ DIY | ❌ DIY | ❌ DIY | | OCI agent distribution | ✅ Built-in | ❌ None | ❌ None | ❌ DIY | | Human-in-the-loop approval | ✅ Built-in | ⚠️ Partial | ⚠️ Partial | ❌ DIY | | CloudEvents audit trail | ✅ Built-in | ❌ None | ❌ None | ❌ DIY | | Multi-channel (WhatsApp, Slack) | ✅ Built-in | ❌ DIY | ❌ DIY | ❌ DIY | | Kubernetes-native | ✅ Built-in | ⚠️ Partial | ❌ DIY | ❌ DIY | | OpenTelemetry observability | ✅ Built-in | ⚠️ Partial | ❌ None | ❌ DIY |

Next steps

Section titled “Next steps”