Build multi-agent AI workflows without rebuilding the plumbing

Forge is our configurable platform for designing, running and governing multi-agent AI workflows - visually with no code, or as low-code definitions. Orchestration, state, retries, tool-calling and the audit trail come built in.

Visual builder & config-as-codeMulti-agent orchestrationGoverned & deploys in your cloud
invoice-triage.forgeno-code ⇄ low-code
New invoiceTrigger
Classify & routeAgent
Extract with PrismTool
Approve over $10kHuman step
Post to ERPOutput
5 nodes·built from Blocks·governed by Sentinel
Build one
The bottleneck

The hard part of an agent workflow was never the agent

A prompt in a notebook is a demo. Turning it into something your business can run - with retries, state, approvals, logging and a way to change it next quarter - is where the weeks go. Forge is built to give those weeks back.

Every workflow rebuilt from scratch

Orchestration, retries, memory, tool-calling, error handling, logging - the same scaffolding gets hand-written for each new use case, then quietly diverges. The interesting 10% ships buried under 90% of plumbing.

Prototypes don't survive production

The demo worked. Then it needed access control, an audit trail, an eval gate, a rollback and someone on call. A script that impressed in a meeting is a long way from a workflow you can put in front of a regulator.

The experts can't build it themselves

The people who know the process best can't touch the code, so every change is a ticket and a wait. Intent gets lost in the handoff between the person who understands the work and the one who ships it.

What Forge does

One platform to compose, govern and ship agent workflows

Forge gives you a visual canvas and a configuration format for the same workflow - build it however suits the person building it, run it the same governed way underneath.

Visual, no-code builder

Assemble a workflow by connecting nodes on a canvas - triggers, agents, tools, branches, human steps. The people who own the process can shape it without waiting on an engineer.

Drag & dropTyped nodesNo engineer needed

Definition-as-code, low-code

Every workflow is also a readable YAML/JSON definition - version it in Git, review it in a pull request, diff it, promote it between environments. Canvas and code stay in sync.

YAML / JSONGit-versionedReviewed in a PR

Multi-agent orchestration

Sequential chains, parallel fan-out, routing, retries, shared state and agent-to-agent handoff are first-class. Compose several specialized agents into one reliable workflow.

Fan-outRoutingRetriesShared state

Bring your own models & tools

Plug in any model - Claude, open-weight, fine-tuned - and any tool or system via MCP, REST or your internal APIs. Forge is the orchestration, not a lock-in to one vendor's model.

MCPREST & internal APIsSwap a model

Human-in-the-loop steps

Drop an approval, a review or an escalation into any workflow as a node. Set the condition - an amount, a confidence score, a risk flag - and a person is brought in exactly when it matters.

Approval gateEscalationLogged decision

Governed & deployable by default

Guardrails, eval gates, versioning, audit trails and observability come standard through Sentinel - and deploy inside your own cloud, not someone else's SaaS.

GuardrailsEval gatesAudit trailYour cloud
See it work

Assemble a workflow, then run it

Pick a workflow, read it as a visual canvas or as its definition, then run it and watch each step execute. This is an illustration of how Forge is built and behaves - not a live model call.

Invoice triage5 nodes
Loading workflow…
workflow: invoice-triage

Press Run workflow to execute these steps and stream the log.

0 steps completed0 ms · audit trail written

Same workflow, two representations - edit it on the canvas or in the definition, deploy it the same governed way.

Under the hood

Six steps from idea to a governed, running workflow

Forge is opinionated about the path to production - the same one whether it's your first workflow or your fiftieth.

1

Compose

Drag nodes onto the canvas or write the definition - triggers, agents, tools, branches and human steps into one flow.

2

Configure

Wire in your models, tools, prompts, data sources and business rules. Set thresholds and the conditions that call in a person.

3

Simulate

Dry-run the workflow against real sample inputs and inspect every step - inputs, outputs, tool calls and cost - before anything is live.

4

Gate

Guardrails and evaluation gates from Sentinel run automatically. A workflow ships only when it clears the bar you set.

5

Deploy

Promote a versioned workflow into your own cloud as an API, a scheduled job or an event handler - rollback is one version away.

6

Observe

Watch runs, latency, cost, success rate and drift in one place - then change the workflow and ship the next version with confidence.

Reference architecture

A configuration layer over your models, tools and data

Forge doesn't replace your models or your systems - it orchestrates them. You build against one runtime; it handles execution, state and governance across everything you plug in.

Build
Buildersvisual no-code canvas · low-code YAML/JSON definition · in sync
Orchestratorsequencing, routing, retries
Forge runtimestate · memory · execution
Registrymodels, tools, prompts, MCP
Built on Blocksplanning · reflection · tool use · ReAct · sequential · human-in-the-loop patterns
Your modelsClaude · open-weight · fine-tuned
Your tools & APIsMCP · REST · internal systems
Your dataDatabricks · warehouse · vector
Deploy targetsAPI endpoint · scheduled job · event handler - in your cloud
Governed end to end - by Sentinel
GuardrailsEvaluation gatesVersioningFull-run audit trailAccess controlObservability & cost
Run

Deployed inside your cloud and your security perimeter - your data and your logic never have to leave to run.

Two ways in, one workflow

No-code for the builder, low-code for the engineer

The disagreement between "give me a canvas" and "give me a file I can review in Git" is a false choice. In Forge they're two views of the same governed workflow - change either, the other follows.

No-codeVisual canvas
New leadTrigger
Enrich & scoreAgent
Score ≥ 70?Branch
Route to sales
Nurture queue

Owned by the person who runs the process. No engineer in the loop to make a change.

in sync
Low-codeDefinition
workflow: lead-qualification
trigger: { on: new_lead }
steps:
- agent: enrich_and_score
model: claude
tools: [crm, web_enrich]
- branch: { if: "score >= 70" }
then: route_to_sales
else: nurture_queue
govern: sentinel/default

Reviewed in a pull request, versioned in Git, promoted between environments like any other code.

In production

The accelerator behind agent workflows we already run

Forge is the platform pattern behind the multi-agent workflows Focaloid ships for clients today - the reason a governed workflow takes weeks, not quarters.

FinTech · PaymentsIn production

Multi-step payment workflows, agent-orchestrated

Payment operations that used to be a chain of manual steps now run as a governed multi-agent workflow - each step validated, retried and logged, with a human brought in only on the exceptions. The composition-and-guardrails pattern that makes that safe is exactly what Forge packages.

Where teams point it next

One platform, many back-office workflows

The same builder gets pointed at invoice and claim triage, lead qualification, KYC and onboarding, support deflection and internal operations. Different logic, different tools - one governed runtime underneath, configured to each.

What teams build on Forge

Wherever a process is more steps than a single prompt

If the work spans several systems, needs judgement at a few points and has to be defensible afterward, it's a Forge workflow.

Finance & operations

Invoice-to-PO matching, statement processing, reconciliation and month-end tasks - orchestrated end to end with approvals where the numbers are large.

Sales & GTM

Lead enrichment and scoring, routing, meeting prep and CRM hygiene - agents doing the research and the busywork, people doing the selling.

Customer support

Triage, intent detection, knowledge-grounded drafting and deflection - with clean escalation to a human, and to Echo for voice, when the case needs it.

Risk & compliance

KYC and onboarding checks, document review, sanction and policy screening - every decision captured with its evidence for the audit that follows.

Data & back office

Enrichment, classification, migration QA and internal-request handling - the long tail of workflows too small to staff and too frequent to keep doing by hand.

Engineering & IT

Ticket triage, incident summarization, code and PR assistance, access requests - internal agent workflows that free the team for the work only they can do.

Why Forge, why us

A platform you own, not a black box you rent

There is no shortage of ways to chain a few prompts together. Fewer are built so you'd run a regulated process on the result - and hand it to your own team afterward.

Configured to you

Your models, your tools, your data, your rules - Forge is the framework, not a template you have to bend your process to fit.

Governed by design

Guardrails, eval gates, versioning and a full-run audit trail come standard through Sentinel - not bolted on when compliance asks.

No-code and low-code

The domain expert and the engineer build the same workflow the way each prefers, and it stays one governed artifact underneath.

Model- and tool-agnostic

Swap a model or add a tool without a rewrite. MCP, REST and internal APIs are first-class, so nothing is welded to one vendor.

Deploys in your cloud

Runs on Databricks, AWS, Azure or GCP inside your security perimeter - your workflows and data stay where they already live.

Yours to keep

We build it with your team and hand it over - definitions in your Git, running in your cloud. No dependence on us to change a workflow.

Solutions & Accelerators

Accelerators that make production faster - and safer.

Two kinds of reusable IP: the tooling we build and govern with, and the solutions that drop straight into a use case. Forge is the platform the first group is assembled and run on.

Forge is the runtime in the middle. Blocks supplies the patterns it composes, Velocity brings the same AI across delivery, Sentinel governs every run - and Prism and Echo are what a Forge workflow looks like once it's pointed at a real use case.

Trust & governance

Every workflow, ready for the review it will face

Forge is built for regulated work. ISO/IEC 27001-certified delivery, governance aligned to US frameworks like the NIST AI RMF and the EU AI Act, an audit trail on every run, and deployment inside your own environment. Governance is applied by Sentinel to every workflow you build - so an agent workflow clears the review instead of stalling in it.

How Sentinel governs it
What's inside

The technology behind Forge

Proven orchestration frameworks and models, assembled into one configurable runtime - chosen for reliability and fit, not for a logo on a slide.

Orchestration & agents

LGLangGraph
CRCrewAI
BLKBlocks patterns
StState & memory

Models & reasoning

CLClaude
OWOpen-weight LLMs
FTFine-tuned models

Tools & integration

MCPMCP servers
APIREST & webhooks
ERPERP / CRM connectors

Data & platform

DBDatabricks
UCUnity Catalog
VECVector store

Deploy & run

AWSAWS
AZAzure
GCPGoogle Cloud
CIGit & CI/CD

Governance

SENSentinel guardrails
AUDAudit & versioning
OBSObservability & cost

The exact stack is chosen per engagement - Forge is framework- and model-agnostic by design, so it can adopt a better tool the day one arrives.

Questions we get

Before you build on Forge

Is Forge a product we buy, or something you build with us?

Both, in sequence. Forge is our platform framework, and we stand up your first workflows with your team on it - then hand it over. The definitions live in your Git and it runs in your cloud, so you're never waiting on us to change a workflow.

How is this different from a general workflow tool or an agent framework?

General workflow tools don't understand agents, models or evaluation; raw agent frameworks give you a library and leave governance, deployment and a visual builder to you. Forge sits in between - the orchestration of a framework, the accessibility of a no-code canvas, and the guardrails, eval gates and audit trail of Sentinel, all in one.

Can non-engineers really build workflows?

Yes - the visual canvas is meant for the person who owns the process. They compose and adjust the flow; engineers set up the harder building blocks and review changes as definitions in a pull request. The canvas and the code are the same workflow, so neither side is boxed out.

Which models and tools does it support?

Any model - Claude, open-weight or your own fine-tuned models - and any tool or system reachable via MCP, REST or an internal API. Forge is the orchestration layer, so you're free to use the right model for each step and change it later without a rewrite.

How does governance actually work?

Every Forge workflow runs inside Sentinel: guardrails on inputs and outputs, evaluation gates before deploy, versioning, access control and a full audit trail on every run. Governance is a property of the runtime, not a checklist someone remembers to apply.

How fast can we get a workflow live?

A focused first workflow - one process, real inputs - is a matter of weeks. Because the orchestration, state, deployment and governance are already there, the work is the logic that's specific to you, not the plumbing that isn't.

Have a workflow worth building once and running for good?

Bring us a process that's more steps than a single prompt. On a 30-minute call we'll map it onto Forge - the nodes, the models, the human steps and the governance - and show you what it takes to get it live.

ISO/IEC 27001-certified · Governed by design · Deploys in your cloud