Skip to main content
Product · Roadmap

What we’re
building.

Active work, committed items and longer-horizon ideas. Priorities move with what people actually run into — this is the current shape of things, not a schedule.

A winding road with signposts at its bends and a flag at the far end
Fig. 20 · The road
Workflows — HubRuns
Workflow run history: counts of active, queued and completed runs with the latest activity
Board · 12 items

Three horizons, no dates.

A quarter printed on a card is a promise nobody made, and it is wrong the moment the quarter turns. The column is the horizon; the tag is the state it is actually in.

Now

In flight4

Faster planning

Lower latency from prompt to first executing branch — plan steps arrive in under 800 ms.

In progress

Interactive previews

Every change gets a live preview environment. Click, inspect, test — no local setup.

Shipping soon

Workflow versioning

Branching and rollback for pipelines. Diff two versions side by side before promoting.

In progress

Advanced run analytics

Per-agent cost breakdown, p95 latency trends, and anomaly alerts you can act on.

In progress

Next

Committed4

Deeper integrations

First-class connectors for Linear, Jira, GitHub Actions, Slack and Notion — not wrappers.

Planned

Team analytics

Usage dashboards per team. Where the time goes, and where it stops.

Planned

Agent marketplace

Publish and discover agent templates. One click to fork and customise.

Planned

Scheduled triggers

Cron-like scheduling for recurring workflows, with timezone support and audit logs.

Planned

Later

Exploring4

Self-hosted compute

Run the agent's miniature servers on your own infrastructure — same API, your VPC.

Exploring

Mobile app

Native iOS and Android — review runs, approve tasks, get notified on the go.

Exploring

Multi-modal agents

Agents that read and produce images, audio and video alongside text.

Exploring

Inter-agent protocol

A standard for autonomous agent-to-agent task delegation across workspaces.

Exploring
How a card moves

Four inputs, no scoring model

These are the things that actually move an item between columns. They are deliberately unweighted: the version of this page that printed a percentage against each one was measuring nothing.

01
What runs, and what fails
Largest input

Task logs are the largest input by a distance. A step that people retry, abandon or work around is a roadmap item whether or not anybody filed it.

02
What you tell us directly
Direct

Support threads and replies. A small user base has one real advantage — a single message can change what gets built next, and often has.

03
What we hit ourselves
Dogfood

The product is built with the product. Most of the Now column started as something that got in our own way first.

04
What has to come first
Ordering

Some items are gated on something unglamorous — a limit, an isolation boundary, a billing shape. Those get done before the thing that needs them, which is why Next is not sorted by how interesting it is.

The other half of a roadmap

What shipped is dated. What is planned is not.

A roadmap is a claim about the future and should be read as one. The changelog is the part with evidence in it: what went out, when, and what broke — including the entries nobody would have chosen to publish.

DatedIncludes regressionsNo marketing pass
Operator — QuoteAwaiting
A quote the agent priced from the rate card, held for approval before it is sent

Push something onto the board.

The fastest way onto the Now column is to run into the thing that is missing and tell us.

A winding road with signposts at its bends and a flag at the far end
Fig. 20 · The road