CALIBER
Quickstart
MLflow-integrated · Agentic workflow control plane

CALIBER : Contextual Adaptive Lifecycle for Intelligent Build, Evaluation, and Refinement

CALIBER is an MLflow-integrated ASGI control plane and application for teams shipping LLM agents to production. Author prompts, tools, skills, MCP servers, knowledge bases, and visual workflows — then test, calibrate, deploy, and observe them through asset-specific controls. Where eval-only tools stop at a score, CALIBER connects measured prompt proposals to explicit review and apply, while each other asset family exposes the lifecycle controls it actually implements.

5-minute quickstart Layered architecture Choose your path Browse docs Examples
Author & compose Versioned prompts, subprocess-contained tools, skills, MCP servers, and a visual Workflow Studio.
Verify & calibrate Test sets, fail-closed scorecards, custom LLM judges, and asset-specific calibration.
Govern & observe Runtime approvals, RBAC, durable prompt-release recovery, MLflow tracing, and the embedded Aria copilot.

Choose your path

Start from your job, not from the repository structure. These entry points route to the shortest useful path for integration, operations, trust review, architecture, or product fit.

Start here
All reader paths The role-first entry page for developers, operators, evaluators, architects, and decision-makers.
Developer
Build & integrate Start with the SDK, then move into auth, scoping, REST, and exact API reference.
Operator
Operate CALIBER Bring the system up locally, verify readiness, and then move into observability and recovery.
Governance
Trust and governance Follow the path across test sets, evaluation, calibration, QA, and apply/release semantics.
Architecture
Deep system understanding Use the curated architecture guide to jump to the right deep page instead of reading the whole tree.
Strategy
Does CALIBER fit? Start from the concise decision-maker overview, then inspect market position, roadmap, and technical proof.

Design principles

Ten principles govern what CALIBER builds and what it refuses to build. The order is deliberate, and when two principles pull against each other, the lower number wins. The full explanation of how the platform discharges each principle lives in the layered architecture.

  1. Governed Agentic Workflows
    Governance, security, compliance, and policy enforcement at the core.
  2. Progressive Autonomy
    The path from human-led processes to fully autonomous agent-driven execution.
  3. Auditability & Observability
    Every workflow, decision, and agent action traceable, transparent, and explainable.
  4. Evaluation by Design
    Testing, verification, validation, benchmarking, calibration, and optimization from the start.
  5. Open & Extensible
    Enterprise systems, AI models, tools, and external agent platforms through open APIs and SDKs.
  6. Modular & Composable
    Loosely coupled components that are reusable, flexible, and easy to extend.
  7. Scalable & Reliable
    Enterprise-grade performance, resilience, and operational reliability.
  8. Open-Source First
    Proven open-source building blocks with strong community support.
  9. Developer & User Friendly
    A clear, productive experience for both developers and business users.
  10. Future Ready
    Evolving AI capabilities, orchestration patterns, and enterprise needs.

Browse by documentation area

The primary documentation areas mirror the published information architecture: start here, use the product, build and integrate, operate safely, inspect exact reference, and then drop into architecture or strategy when you need deeper context.

Start here

Orientation, role-based entry points, and the shortest path to a useful starting page.

Start here Overview Product overview, quickstart entry, and the main documentation landing page. Start here 5-minute quickstart The shortest path to a real CALIBER success: local bring-up, deployment verification, and the right next guide. Start here Choose your CALIBER path The role-first entry page for developers, operators, evaluators, architects, and decision-makers. Start here Decision-maker overview The concise non-implementation path through what CALIBER is, what it solves, and where to go next.

Use CALIBER

Feature areas, trust surfaces, and the pages that help system users and evaluators understand product behavior.

Use CALIBER Workflows The practical path for authoring, running, deploying, and debugging workflows. Use CALIBER Prompts The task-oriented path for prompt authoring, evaluation, calibration, and release. Use CALIBER Knowledge bases The practical guide to ingesting source material, improving retrieval, and grounding assistant behavior. Use CALIBER Aria assistant The supervised assistant path for sessions, tool use, approvals, and review boundaries. Use CALIBER Trust and governance The curated path across test sets, evaluation, calibration, QA, and apply/release controls.

Build & integrate

Programmatic integration, SDK usage, raw HTTP guidance, and the development paths that get developers to a working integration quickly.

Build & integrate SDK guide Install, authenticate, scope to a project, and make the first CALIBER call from Python. Build & integrate Auth and project scoping The minimum auth and scoping decisions a developer must get right before calling CALIBER programmatically. Build & integrate SDK vs REST API The first integration decision for developers: typed Python surface or raw HTTP. Build & integrate Error handling and retries The practical integration policy for classifying failures, using waiters, and retrying safely. Build & integrate CI/CD automation The repository-grounded automation path for docs, tests, and integration-safe delivery. Build & integrate Developer troubleshooting The first-pass diagnosis path across auth, project scoping, CSRF, uploads, waiters, and workflow invocation.

Operate CALIBER

Bring-up, configuration, readiness, storage, troubleshooting, and recovery paths for operators and admins.

Operate CALIBER Local bring-up The operator-first path for starting CALIBER locally and verifying readiness. Operate CALIBER Configuration and provider setup Ports, database, provider, gateway, object-store, workflow-storage, and event-backend setup. Operate CALIBER Health and readiness The operational meaning of CALIBER liveness, readiness, and runtime evidence. Operate CALIBER Operations runbook The on-call recovery guide for failures CALIBER cannot settle on its own.

Examples

Start here when you want complete runnable flows instead of individual API or architecture pages.

Examples SDK recipes Full runnable examples that use only the Python SDK plus the standard library. Examples Cookbook gallery The platform cookbook catalog with every scenario, readiness requirement, and installable recipe.

Reference

Exact API, SDK, workflow-component, and configuration lookup material.

Reference SDK API reference Typed client entry points, resource modules, models, errors, waiters, and async support. Reference Resource catalog The route families grouped by the domains developers actually build with. Reference HTTP reference The concrete contract map for representative operations and the published OpenAPI entry points. Reference Workflow components The built-in workflow node catalog and the component model behind visual workflow authoring. Reference Config and environment reference The fast index for the repository’s current configuration families and representative environment variables.

Architecture

Deep architecture, runtime boundaries, and the technical model behind the platform.

Architecture Architecture reader guide The curated entry point into the architecture corpus, organized by technical question. Architecture Layered architecture The six-layer stack, lifecycle model, trust boundaries, and extension seams. Architecture Interactive layered architecture A presentation-grade explorer for tracing one layer, one path, or one node at a time. Architecture Platform Embedded and standalone topologies, shared runtime state, async workers, and trust boundary. Architecture The refinement loop The canonical prompt-refinement path and the review, alias, and rollback boundaries that differ across asset families.

Strategy

Market context, roadmap, and the supporting material used to explain where CALIBER fits.

Strategy Competitive analysis CALIBER vs. Langflow/Flowise/Dify, n8n/Flowable, the LLMOps & eval tools, MLflow GenAI, and the AWS/Google/Microsoft stacks. Strategy Roadmap The quarter-by-quarter plan derived from the analysis and verified against the architecture.

First success

Pick the shortest path that proves CALIBER is working for your job. These are the pages to open first for local bring-up, SDK integration, trust review, and workflow authoring.

Run CALIBER locally

Use the shortest bring-up path, confirm readiness, and move into operator guides only if you need them.

Start with the 5-minute quickstart

Make the first SDK call

Start with the typed client, then drop into auth, scoping, or raw HTTP only when integration needs it.

Open the SDK guide

Review trust posture

Follow the path across evidence, evaluation, calibration, QA, and apply or release boundaries.

Open trust and governance

Build a workflow

Start from the practical workflow guide before you read deeper runtime or architecture pages.

Open workflows

Secondary resources

These pages stay published because they help with walkthroughs, demos, and presentations, but they are intentionally secondary to the main documentation paths above.

Interactive and hands-on

Visual and guided material that complements the main guides without replacing them.

Operate CALIBER Guided walkthrough A step-by-step operator walkthrough for a local product tour and initial system verification. Architecture Interactive layered architecture A visual exploration layer for the platform architecture and subsystem boundaries.

Collateral

Presentation-grade summaries for sharing CALIBER outside the core documentation flow.

Collateral Product presentation The main presentation surface for demonstrating CALIBER in a browser-friendly narrative format. Collateral Timed presentation The timed or speaker-paced presentation variant published as secondary collateral.