The software factory where
people and agents ship together.

Decisions, codebase context and coding agents in one stack, under one model policy.

$ curl -fsSL https://latentstack.dev/install.sh | bash

Installs the LatentCode CLI, the LatentWork desktop app and the editor extension.

Everything a team needs to ship with agents. Use what you need. It all works together.

01

Workspace

Talk it through with your team and agents. Every decision lands in a living wiki that agents read before they start.

Open the workspace

02

LatentGraph

Ask what any change touches. Agents get your architecture, dependencies and hidden couplings over MCP.

Explore lgraph.dev

03

Managed agents

Start agents from the terminal or the app. Run a hive locally, or hand long jobs to agents that keep going in the cloud.

See how hives run

04

Console

Set models, guardrails, budgets and access for every team. The gateway enforces them on every call.

Choose where it runs

Every codebase has a hidden graph. LatentGraph makes it visible to your agents.

Explore lgraph.dev

Architecture, dependencies, intent, and the runtime couplings static analysis misses, mapped from your code and PR history. Any MCP coding agent can query it.

Humans add contextAnnotate intent, add missing edges, attach specs.
Agents use and enrich itEach session can discover new edges and write them back.
Owners approveNothing changes without owner approval.
The graph growsFor the whole team. And you own it.
$ npm install -g @latentforce/latentgraph $ lgraph start $ lgraph init

The free version indexes on LatentForce servers. On-prem and air gapped deployments index inside your environment.

What touches reports.export?
Analyzing: reports-service 100%
ui/api/auth/events/reports/lib/timedb/billing/requireRole()exportCsv()audit events
CoreModuleServiceUtilPrimaryImplicit

Better context. Measurably fewer wasted runs.

Read the evidence

66%

lower API cost per resolved task for Claude Code with LatentGraph.

Without$3.93
With$1.33

10/14

hard tasks resolved with it, against 8 of 14 without.

Without8
With10

8/10

PR tasks completed in a healthtech customer evaluation, against 2 of 10.

Without2
With8

Choose where it runs. You decide where the data goes.

Managed

We run it for you. Your team connects in, and model calls go only to the providers you approve.

Runs atLatentForce
ModelsPrivate + approved
What can leavePermitted inputs only

Your infrastructure

Bring your own cloud: your account, a GPU cloud, or your own servers. Nothing leaves unless you allow it.

Runs atYour team
ModelsPrivate + approved
What can leaveOnly what you allow

Air gapped

Fully isolated. Runs on local models, with no network path out.

Runs atYour team
ModelsLocal only
What can leaveNothing

Ship with agents.Keep control.

Install in one command. Connect a repo. Ship your first change today, and start with just the component you need.

$ curl -fsSL https://latentstack.dev/install.sh | bash
Request an evaluation