Fennwick Hale
@fennwick
ModeratorMODERATOR
Leader · Member since Mar 2025
Moderator. Vault of 41,000 notes and counting.
Location: Leeds
0 followers · 0 following
- Topics created
- 0
- Posts created
- 0
- Likes given
- 29
- Likes received
- 0
- Days visited
- 0
- Solutions
- 0
Days visited count across the whole community, not just this forum.
Most active topics
Nothing yet.
Most liked posts
Nothing yet.
Top categories
Nothing yet.
Top tags
Nothing yet.
- The API reference should show the response headers
Agreed — this is now on the reference roadmap.
Likes given · 10/12/2026
- The API reference should show the response headers
Strong agree. I learned about x-request-id from this forum, not the docs.
Likes given · 10/12/2026
- The API reference should show the response headers
Rate-limit headers, request ids, and the processing-time header are all in every response, but the reference only documents the body. Half the questions in this
Likes given · 10/12/2026
- Handing off between two agents — shared memory or pass everything?
Summary plus the last three user messages verbatim. The summary carries facts, the verbatim messages carry tone.
Likes given · 10/12/2026
- Handing off between two agents — shared memory or pass everything?
I have a triage agent and a refund agent. When triage hands off, do people pass the full transcript or a summary? Full transcript feels wasteful, summary feels
Likes given · 10/12/2026
- Tracking per-user cost in a multi-tenant app
Same. Also tag requests with a feature name in metadata, so you know which feature the expensive users love.
Likes given · 10/12/2026
- Tracking per-user cost in a multi-tenant app
We pass our user id in the user field and join the usage export on it nightly. Takes ten lines of SQL.
Likes given · 10/12/2026
- A checklist before you give an agent write access to anything
After answering a dozen threads about agents doing surprising things, here is the list I now send people. Dry-run mode. Every write tool has a dry_run flag that
Likes given · 10/12/2026
- Vantor Builders Meetup — Berlin, 14 November
We are hosting an evening for developers in Berlin: three short talks, demos from the community, and pizza. When: 14 November, 18:30 Where: a canal-side venue i
Likes given · 10/12/2026
- Temperature 0 still gives different answers — is that expected?
Zero means greedy, not deterministic. Batched inference on GPUs can change floating-point order between runs, which occasionally flips a near-tie between two to
Likes given · 10/12/2026
- Please let us pin conversations in the sidebar
I have three conversations I return to every day and twenty I opened once. A pin would save a lot of scrolling. Anyone else?
Likes given · 10/12/2026
- Structured output: how do I allow an optional field?
Works. Slightly odd, but works.
Likes given · 10/12/2026
- I built a recipe app that reads your fridge — feedback welcome
Point the camera at your fridge, it lists what it sees, and suggests three dinners. Built with Halo 2 vision and about 600 lines of Swift. App screenshot What I
Likes given · 10/12/2026
- Which embedding dimension should I use for 2M short product titles?
Short titles do not carry enough information to fill 3072 dimensions. I would start at 1024 and only go up if you can measure a gap on your queries.
Likes given · 10/12/2026
- Which embedding dimension should I use for 2M short product titles?
We tested on 500k SKU titles. Recall@10: 256 → 0.81, 1024 → 0.86, 3072 → 0.87. The jump from 1024 to 3072 was not worth 3× the storage for us.
Likes given · 10/12/2026
- Which embedding dimension should I use for 2M short product titles?
We are embedding about two million product titles (5–15 words each) for semantic search. vx-embed-3 supports 256, 1024 and 3072 dimensions. Storage is a concern
Likes given · 10/12/2026
- Share your shortest system prompt that actually changed the output
Answer, then explain. Stops it from warming up for three paragraphs.
Likes given · 10/12/2026
- Project files disappear from Vantor Chat after a day
Same here on Windows. The files come back if I sign out and in again, so it looks like a display problem, not deletion.
Likes given · 10/12/2026
- How we cut our agent's tool-call loop from 9 turns to 3
Did trimming tool output change accuracy at all? I worry about the model missing a field it needed. Accuracy went up slightly (94.1% → 95.3% on our 400-booking
Likes given · 10/12/2026