[ RM-01 ]
Multi-run traces
One trace view across a chain of runs — see where context flowed and where it didn't.
Platform — Roadmap/roadmap
The same sequence discipline we apply to the company, applied to the codebase. No dates — order is the promise, not deadlines.
[ 01 ]The board — read left to right
Now
in active development[ RM-02 ]
Latency p99 work
Tail latency on run start. The median is fine; the slowest 1% is not, and that's what people feel.
[ RM-03 ]✓ shipped
Desk v1The two-way ask desk at /help/ask — questions in, answers back, threads kept.
Next
designed, not yet started[ RM-04 ]
Context graph permissions UI
See and edit exactly what a run may read, per source — the trust kernel made visible.
[ RM-05 ]
SDK v2
Streaming-first clients with typed traces, retiring the v1 request/poll loop.
[ RM-06 ]
World expansion
More dioramas for /world — real places where the product quietly works, not sci-fi.
Later
deliberate, not yet earned[ RM-07 ]
Second lane exploration
Only after the wedge proves retention. The sequence is the strategy — we don't skip ahead.
[ RM-08 ]
On-prem runtime
Y0 inside your perimeter for regulated workloads. Real demand exists; so does real cost.
Items move right to left only when the column before them is genuinely done. "Later" is not a graveyard — it's a commitment to not start things we can't finish. Shipped work is recorded in the changelog.