Compass · production AI platform

Build AI that
does real work

Build agents and workflows on your own models, data, and APIs — then ship them as immutable, versioned deployments and trace every execution, step by step. The production depth that automation tools skip.

Get startedExplore the platform
input
Type
Chat
Output
Database Retrieval Agent
Connection
OpenAI · gpt-5.1
Input
input
System Instructions
Query the database and return the records most relevant to the user’s request.
Tools
postgres
Agent Configuration
Configure...
Response
File Analyzer Agent
Connection
Anthropic · claude-opus-4.5
Input
Database Retrieval Agent
System Instructions
Analyze the retrieved files and produce a structured summary of the findings.
Tools
s3
Agent Configuration
Configure...
Response
Output
Single
Agent
File Analyzer Agent
✎ Edit output
postgres
Connection
primary-db · Postgres
Tool
s3
Connection
analytics-lake · S3
Tool

Connects to your stack

AnthropicOpenAILM StudioPostgresApigeeOpenAPIREST APIsAnthropicOpenAILM StudioPostgresApigeeOpenAPIREST APIs
Beyond automation

Connecting a model to a workflow is the easy part. Compass is built for everything after the demo — immutable deployments your team can run, human sign-off on the actions that matter, and a trace of every execution, down to the token.

Read the docs
The building blocks

One platform, from first connection to live run

Compass gives your organization a single place to build AI that does real work — and to operate it once it’s live.

Connect

Providers, integrations & gateways

Three kinds of connection, not one: model providers — Anthropic, OpenAI, even local models via LM Studio — integrations to your data and APIs (Postgres, or any OpenAPI spec as typed actions), and gateways like Apigee to route and govern model calls.

Build

Agents & workflows, as peers

Give an agent instructions, a model, and tools — or wire a multi-step workflow on the canvas. Neither is the advanced one: agents call workflows, workflows contain agents, and agents delegate to other agents.

Deploy

Immutable, versioned releases

Deploy freezes a draft into a numbered version that never changes, then points the live deployment at it. Roll back to any earlier version instantly and atomically. Your team runs only what you’ve shipped.

Observe

Executions are traces, not logs

Every execution records its input, output, tokens, and every step it took — nested agent and tool calls you can expand inline. Watch it stream live, or replay node by node. Latency, success rate, and usage roll up to the whole project.

Built for production

Where most AI tools stop, Compass starts

Shipping AI your team depends on takes more than wiring boxes together — it takes versioned releases, human control over the actions that matter, and a full record of everything that runs.

Deploy

Ship versions you can trust

Every deploy freezes the draft into an immutable, numbered version — versions never change.
A version captures the exact versions of the agents and workflows it depends on, so it runs the same way every time.
Roll back to any earlier version instantly and atomically; runs already in flight are untouched.
Drafts stay editable while your team and systems run only what you’ve shipped.
Govern

Keep humans in the loop

Mark any tool, action, or agent as requiring approval — the execution pauses until someone signs off, then resumes exactly where it stopped.
A rejection isn’t a crash: the agent is told no and adapts, and every decision is recorded — who approved what, and when.
Spend limits on an organization, project, or connection stop a runaway workflow before the model is even called.
Roles separate who builds, who runs, and who only watches — down to read-only viewers.
Observe

Trace every execution, end to end

Every execution is a trace — its input, output, tokens, and every step, not just a log line.
When an agent calls another agent or workflow, expand its trace inline, all the way down.
Watch tool calls and output stream in live, or replay a finished execution node by node.
Tokens and p95 latency roll up from a single step to the whole project; success rate flags when it dips.
For developersBuild in the UI, or drive the whole lifecycle from code.
Async Python SDKFull CLIREST APILive SSE streamingScoped API keys
The runtime

Built to run at scale

Every agent and workflow runs on a durable execution engine — the layer most AI platforms don’t have.

1,248Active now
99.9%Success
1.2sp95

Durable

A failed step or a restarted machine resumes from its last checkpoint — never from scratch.

Scalable

Thousands of concurrent executions, batch fan-out over whole lists, long-running jobs, and nested sub-workflows — orchestrated reliably.

Isolated

Each execution runs against its exact deployed version, on its own, with automatic retries on transient failures.

Solutions

The platform is the foundation

Some teams build on Compass themselves. For everyone else, there are solutions — complete AI applications that Sage designs, builds, and operates for you, on this same platform.

Built by Sage

You bring the process

Our engineers scope the work with your team, compose the agents and workflows, and wrap them in an application made for the people who'll use it. Nothing for you to build.

On Compass

The same platform underneath

A solution ships like everything else here — versioned releases on the durable runtime, hosted and operated by Sage. Your team just signs in.

Glass box

You see everything it does

Your team gets access to the project behind the solution — the workflows, the agents, and a trace of every run. Custom AI, without the black box.

How we build solutions

Put your AI to work

Build, deploy, and operate AI agents and workflows for your organization — with the governance and visibility that production demands.

Get startedBook a demo