OXYGENOXYGEN

For GTM engineers

The hosted runtime GTM engineers were missing

You already know how to build the motion. What you lack is somewhere to run it that keeps state, retries and provenance for you.

try

A GTM engineer's job is glue. A scraper here, an enrichment vendor there, a sheet holding the join, an automation tool firing webhooks at three in the morning, and a sequencer that knows none of it. The logic was never the hard part. The glue is.

Oxygen is the runtime that glue stood in for. Tables give you typed columns, waterfalls and state on every cell. Workflows give you triggers, branches, retries, approval checkpoints and durable runs on a Postgres queue. Records hold whatever survives the working list.

Who this is for
The person who builds the GTM machinery at a four to fifty person B2B SaaS company: an in-house GTM engineer, a technical founder, or a fractional operator brought in to make it repeatable.
What it looks like today
The stack is six vendors and duct tape. A scraper feeds a sheet, an automation tool moves rows, an API bills per attempt, and when something breaks overnight there is no run to open, only a failed webhook.
What changes
The plumbing becomes the product. Typed tables, a column runtime, a Postgres-backed queue, versioned workflows and a readable run log ship as primitives you did not have to write or babysit.

Example searches

Ask for it the way you'd say it

Every search below runs on the same hosted Tables, with the cost previewed before anything is spent.

  • Rebuild a waterfall you own

    take my clay table export and rebuild these columns so the waterfall and its runs live in one place

    Each column keeps its inputs, cost and error state, so a failed row becomes a retry.

  • Stack and hiring on one row

    find saas companies that opened a revops role this quarter and detect what crm they run

    The posting and the detected technology arrive as separate typed columns you can rerun.

  • Build it from the terminal

    give me the exact commands to build this list, enrich it and check what the run cost

    Capability search returns the owning command, then a schema call prints inputs and cost.

Plays

Three motions you can run this week

Each one is a chain of Oxygen primitives — the same hosted objects your workspace already has, composed.

  • Move the glue into hosted columns

    Every column carries its own inputs, cost, provenance and retry.

    1. 1Add formula, AI and tool columns, then run a few rows first.
    2. 2Promote what clears your bar and open the run to read the cost.
    oxygen columns add
  • Deterministic orchestration, real retries

    A versioned workflow with triggers, branches and approval checkpoints.

    1. 1Author the definition in files, lint it, then apply the hosted version.
    2. 2Subscribe it to an event type, then tail the run that failed.
    oxygen workflows apply
  • Publish the motion as a tool

    A workflow your assistant calls by name, under the same approvals.

    1. 1Turn on MCP exposure so clients see it as a callable tool.
    2. 2Give an agent a bounded version and an approved credit ceiling.
    oxygen workflows mcp enable

Capabilities

What you get

  • A queue you did not build

    Long work becomes a durable run on tenant Postgres with leases, heartbeats and attempt budgets. Nothing depends on your laptop staying awake.

  • Provenance down to the cell

    A cell keeps its inputs, provider, cost, error and history, so you can answer why a value is what it is without paying to rerun it.

  • Three surfaces, one contract

    Web, CLI and MCP expose the same customer actions over the same API, so a motion you script is one a teammate can operate.

  • Spend bounded on purpose

    Paid steps preview scope and cost and execute under a ceiling. Tool authority and spend authority are separate, so a trigger cannot widen its own budget.

Instead of

The stack this replaces

Native, on one contract and one credit balance — not another tab wired to the last one.

  • Clay
  • n8n
  • Zapier
  • Airtable
  • Apollo
  • Smartlead

Data sources

What the data actually comes from

Every value lands with its provider and cost recorded on the cell.

  • People Data Labs

    Firmographic and person records at the volume a list build actually needs.

  • TheirStack

    Job postings as a queryable filter, so a hiring decision becomes a column.

  • Firecrawl

    Crawls and extracts pages, turning a prospect site into input for an AI column.

  • Apify

    Actor-based scraping for public sources with no API, inside the same queue.

  • Blitz API

    Heads the work-email chain, keeping the metered lanes behind it idle.

Run these on Oxygen's managed credits, or connect your own provider keys and pay the vendor directly — the same columns, the same runs, the same provenance either way. See every integration.

Limits

Where this stops

  • Hosted means hosted. You can author a workflow in files and apply it, but a step cannot execute arbitrary code you upload. An opaque runtime would break the approval and provenance model.

FAQ

Questions people ask first

Can I keep workflow definitions in git?
Yes. Local files are authoring input: write the schema, lint it, apply it. The hosted version is what executes, and export pulls it back out for review.
What happens when a step fails overnight?
The run stays durable. Steps carry attempt budgets, leases and heartbeats on a Postgres queue, so a worker restart resumes instead of dropping the work.
Is this a spreadsheet with AI columns bolted on?
No. A table is a typed dataset with per-cell action state, a run history per column, and row-level binding to canonical records. The grid is only the view.
How does an assistant call a motion I built?
Through the MCP server. A workflow can be exposed as a callable tool with a schema. Paid steps still preview scope and need an approved ceiling first.
Do I lose my own provider keys?
No. Managed access is the default, and a lane that supports a key can draw your own vendor balance instead. Bringing your own keys is a paid-plan capability, and the run log names which lane resolved each row either way.
What is free here and what draws credits?
Tables, records, workflows, agents and sequence drafts are unmetered. Credits go to enrichment, AI columns, provider signals and inference, each previewed first.

Put your GTM motion on one stack

Sign up, get a working workspace with a one-time credit grant, and run the first play end to end without wiring five tools together.