Roadmap
What exists today for each North Star, platform work in flight, capabilities not yet built, and what is parked from the pilot. Direction and priorities are the owners'.
On this page
TL;DR: One thing now: the dev factory. Each North Star has an owner role who sets its direction, priorities, milestones and targets. This page states what exists, what the platform team is building to support it, and what is parked. No dates are promised.
Now: the dev factory
| North Star | What exists | Direction |
|---|---|---|
| Agentic flows | One workflow, core-dev, from issue to pull request with three human decisions, run so far against a stand-in model. |
The agentic-flows owner's. |
| Rich context | The brain, recall, Drive and the connectors, mostly empty or switched off. | The rich-context owner's. |
| Measurement and reporting | Detailed cost and activity records, reports, exports and long-term history. Nothing about value. | The measurement-and-reporting owner's. |
Platform work under way
| Item | State |
|---|---|
| Slim production base: Vertex only, clean schema, two deployment shapes on GCP | Done; each cloud shape proven once, then taken down |
| Long-term history in ClickHouse with verified purge | Done |
| Connectors: Salesforce, Atlassian with Jira events, Google, read-only ClickHouse | Merged; proven against stand-ins; live sign-in needs Astound's credentials |
| Full test suite green | In progress |
| Pilot data loaded as test data | Done, local only |
| Landing page and full documentation for what ships | In progress |
| Real model, GitHub App, single sign-on (AD FS), required CI check | Waiting on Astound's project, credentials and app installs |
Platform capabilities not yet built
Candidates the platform team can build. Whether, and in what order, is for the owners to say.
| Item | For | Size |
|---|---|---|
| GitHub issue trigger | Flows | An event adapter; the webhook entry point exists |
| Tests and CI results fed back into a run | Flows | New step kind or tool, plus CI wiring |
| Review comments on an open PR handled by the agent | Flows | New trigger and loop |
| Tool list enforced per step; retry defect fixed | Flows | Fix in this repository |
| Sandbox per run | Flows | Design needed |
| Governance and gateway scanners from warn to enforce | All | Configuration, after a measurement window |
| Extractors, bulk import, agent proposals, eval harness | Context | One piece each |
| Outcome capture: PR merged, review rounds, reopened | Measurement | GitHub and Jira ingest linked to runs |
| KPI snapshots, targets, scorecard | Measurement | Tables and one report page |
| Attribution and derived facts exported to ClickHouse | Measurement | Two more datasets in the export |
| More kinds of work: update, deploy, heal, repair | Flows | One workflow each; not designed |
Future pipeline: parked from the pilot
Removed from this repository unless marked otherwise.
| Feature | To bring back |
|---|---|
| Desktop bridge (Claude Desktop and Cowork connection) | Restore the client, packaging, signing and release pipeline; the server side is still in core |
| Wider marketplaces, plugins and skills | Content only; today one general marketplace plus the connectors marketplace ship |
Business workflows beside core-dev |
Content, plus the fact kinds and brain profiles removed with them |
| Kits and remote sources | Restore the export tool and bundle rules |
| Catalog and share by link | Restore one crate from the pilot |
| Wider analytics (analysis section, conversation judging, more lists and panels) | Restore pages and jobs |
| Other model providers beside Vertex | Configuration and Astound's sign-off; core still supports them |
| Slack and Teams | A core feature to re-enable |
| Knowledge bank | Drive and the brain cover part of it |
| Single sign-on (AD FS) | Kept, switched off; needs Astound's trust configuration |
| Other hosting (Fly, GitHub Actions release) | Replaced by GCP and Cloud Build |
Developers using Claude Code through the gateway with a personal access token works today and is kept.
How something leaves the pipeline
An owner asks for it and says what it is for. The platform team sizes it and says what it displaces.