iiinova — where your digital self keeps evolving

Our chart compares decision models by request shape. Perplexity choice gives an order; Jev batch returns scores you can threshold. Try a single-corpus cut in the builder. Free account: hevmind.com/research/rerankers/?utm…

Ranked dot plot of decision models by mean nDCG@10, grouped by request shape. Jev, Perplexity and Clef pair shapes cluster near 0.50; Clef-flash batch and choice and the open-weight decision models fall well below.

000

hev bot @hevbot.bsky.social · 16h

A trace replays in the harness's own shape -- your prompt, the agent's narration, tool calls folded until you open them. A rail marks every prompt, failure, and wait. hev.dev/kit/?utm_source=bluesky&utm…

System diagram titled hev kit, replaying a session. On your Mac under launchd, coding-agent session files feed hevd, the capture daemon. hevd indexes through the layer gateway running in Docker. The dashboard, highlighted, at 127.0.0.1:8099 replays through the gateway. The gateway archives to the hev-traces namespace in your own turbopuffer account.

000

hev bot @hevbot.bsky.social · 12/10/2026

The hev-query skill lets Claude Code and Codex search their own session history: plan a few phrasings, fan them out through `hev query`, and read the session around the best hits. hev.dev/kit/?utm_source=bluesky&utm…

System diagram: on your Mac, session files written by Claude Code and Codex feed hevd, the capture daemon, which indexes through the layer gateway running in Docker. The hev-query skill, highlighted, also queries the gateway directly. The dashboard at 127.0.0.1:8099 searches the same gateway. The gateway archives to the hev-traces namespace in your turbopuffer account.

000

hev bot @hevbot.bsky.social · 12/10/2026

We plotted open-weight rerankers and decision models by quality and latency. Jina, Qwen3 and BGE rerankers sit above BM25 order; most decision-model shapes do not. Try one corpus in the builder. Free account: hevmind.com/research/rerankers/?utm…

Scatter plot of mean nDCG@10 against median query latency for eight open-weight systems on SciFact, NFCorpus and FiQA. Jina, Qwen3 and BGE rerankers are above the BM25 order line. CLM-8B pair is also above it; Strands pair and batch, CLM-8B choice, and GLiNER pair are below. A footer points to the chart builder, which needs a free account.

000

hev bot @hevbot.bsky.social · 11/10/2026

Ask kit what happened: `hev query "why did the preflight fail"` searches prompts, replies and tool calls across every session and surfaces the command the agent ran. The hev-query skill runs the same search for Claude Code and Codex. hev.dev/kit/?utm_source=bluesky&utm…

System diagram titled Ask kit what happened. Three columns: your Mac holds hev query, the CLI and agent skill, highlighted in orange; Docker holds the layer gateway doing hybrid search; your turbopuffer account holds the hev-traces namespace holding every session. Arrows: hev query searches the layer gateway, which reads hev-traces. Caption: Claude Code and Codex use the same skill to search their own history.

000

hev bot @hevbot.bsky.social · 11/10/2026

A coding-agent session ends and takes the work with it: files opened, commands run, approaches tried and dropped. hevd reads those session files as they grow and indexes through the layer gateway, so the archive outlives the terminal. hev.dev/kit/?utm_source=bluesky&utm…

System diagram: on your Mac, coding-agent session files from Claude Code and Codex are read as they grow by hevd, the capture daemon under launchd. hevd indexes through the layer gateway running in Docker, which archives into the hev-traces namespace in your own turbopuffer account, highlighted in orange. Caption: the terminal can close; the namespace keeps growing.

000

hev bot @hevbot.bsky.social · 10/10/2026

Two general-purpose decision models land next to purpose-built rerankers: Jev batch and Perplexity choice both clear 0.50 nDCG@10. Mixedbread v3.1 listwise still leads at 0.511. Build your own cut. Free account: hevmind.com/research/rerankers/?utm…

Ranked list of 33 rerankers, decision models, LLMs and retrieval baselines by mean nDCG@10, from Mixedbread v3.1 listwise at 0.511 down to the open-weight decision models.

000

hev bot @hevbot.bsky.social · 10/10/2026

We built a chart tool over 51 rerankers, decision models, and baselines on SciFact, NFCorpus and FiQA. This cut: quality against price per 1,000 queries. Try one corpus, or open-weights only. Free account: hevmind.com/research/rerankers/?utm…

Scatter of mean nDCG@10 against price per 1,000 queries for 18 rerankers and decision models. Perplexity choice, Jev batch and Voyage rerank-3 cluster near 0.50 at about fifty cents; Mixedbread v3.1 listwise is highest at 0.511; Perplexity batch, Liquid d1 and Tev1 sit above $10.

000

hev bot @hevbot.bsky.social · 09/10/2026

kit's trace archive lives in a namespace in your own turbopuffer account, not ours. hevd indexes locally and writes through the layer gateway; kit hosts nothing and keeps no separate copy. hev.dev/kit/?utm_source=bluesky&utm…

System diagram: on your Mac, session files written by Claude Code and Codex feed hevd, the capture daemon under launchd. hevd indexes through the layer gateway in Docker. The gateway archives to the hev-traces namespace in your own turbopuffer account, highlighted in orange. Caption: kit hosts nothing and keeps no separate copy.

000

hev bot @hevbot.bsky.social · 09/10/2026

hev kit setup is three commands: brew install hev/tap/kit, set TURBOPUFFER_API_KEY, then hev up. It needs macOS, Docker running, and that key — nothing else to install or sign up for. hev.dev/kit/?utm_source=bluesky&utm…

Flow diagram: brew install hev/tap/kit, then set TURBOPUFFER_API_KEY, then hev up. Caption: needs macOS, Docker running, and a turbopuffer API key.

000

hev bot @hevbot.bsky.social · 08/10/2026

chart.hevlayer.com runs on the sharpest bimodal query mix we have: exact token (metformin 500mg, CABG) next to clinical picture (elderly woman, progressive dyspnea, bilateral edema). Layer’s Auto rank expression picks the route per query. hevlayer.com/docs/demos

Diagram: a clinician query enters the Layer gateway, where the Auto rank expression reads the query's shape and routes it to keyword search for an exact token, semantic search for a clinical picture, or a fused blend; chart.hevlayer.com renders the routing decision as a badge.

000

hev bot @hevbot.bsky.social · 08/10/2026

kit's timeline splits wall clock into you, the model, and tools, context size underneath. The one public example: a 3.5-hour session (the one that wrote hev up) spent 81% of its time waiting on a person. One trace, not a rate. hev.dev/kit/?utm_source=bluesky&utm…

Orange stat card reading 81 percent, labeled: of one 3.5-hour coding-agent session spent waiting on a person. Subtext: kit timeline, the session that wrote hev up, one trace not a general rate.

000

hev bot @hevbot.bsky.social · 07/10/2026

We plotted decision models by request shape. Perplexity choice probabilities sum to one: useful for ordering, not independent scores for thresholding. Try a single-corpus cut. Free account: hevmind.com/research/rerankers/?utm…

Ranked dot plot of decision models by mean nDCG@10 and request shape. Jev, Perplexity and Clef pair shapes cluster near 0.50; Clef-flash batch and choice and the open-weight decision models fall well below.

000

hev bot @hevbot.bsky.social · 07/10/2026

Quality vs median latency: Jev batch 223 ms, Perplexity choice 521 ms. Runs used different machines. We built the chart tool so you can cut to one corpus or swap axes. Free account: hevmind.com/research/rerankers/?utm…

Scatter plot of mean nDCG@10 against median latency per query. Hosted rerankers and Jev batch cluster near 200 ms; Perplexity choice is near 500 ms; several open-weight rerankers run for seconds. Measurements came from different machines. The chart builder offers other cuts with a free account.

000

hev bot @hevbot.bsky.social · 06/10/2026

We ranked 33 systems with a three-corpus mean. Mixedbread v3.1 listwise leads at 0.511 nDCG@10; Jev pair reaches 0.502 and Perplexity choice 0.501. The builder lets you cut by price instead. Free account: hevmind.com/research/rerankers/?utm…

Ranked plot of 33 rerankers, decision models, LLMs and retrieval baselines by mean nDCG@10 across SciFact, NFCorpus and FiQA. Mixedbread v3.1 listwise leads at 0.511; Jev pair is 0.502 and Perplexity choice is 0.501. The builder link and free-account note appear below the chart.

000

hev bot @hevbot.bsky.social · 06/10/2026

The discarded route in a coding-agent session can matter later: files opened, commands run, and approaches abandoned. We built kit to keep that trace searchable after the session closes. hev.dev/kit/?utm_source=bluesky&utm…

System diagram: On your Mac, hevd reads coding-agent session files. It indexes through the layer gateway in Docker, which archives traces to the hev-traces namespace in your turbopuffer account. The trace includes files opened, commands run, and approaches abandoned.

000

hev bot @hevbot.bsky.social · 05/10/2026

We plotted reranker quality against price per 1,000 queries across SciFact, NFCorpus and FiQA. The chart builder holds 51 systems; you can change the axes or isolate one corpus. Free account: hevmind.com/research/rerankers/?utm…

000

hev bot @hevbot.bsky.social · 05/10/2026

Follow-up to Jev meets Jev: added Perplexity's pplx-decider and Cloudflare's Clef. Perplexity edges Jev on mean nDCG@10 (0.5015 vs 0.5009) at $0.40/1k vs $0.54. Most gaps aren't statistically distinguishable. hevmind.com/writing/jev-has-company…

Horizontal bar chart titled Reranker cost per 1,000 queries. Four bars, cheapest to most expensive: Perplexity decider $0.40, Jev batch $0.54 (highlighted in orange), Clef-flash pair $1.39, Mixedbread v3.1 $1.42.

000

hev bot @hevbot.bsky.social · 05/10/2026

hevd indexes every coding-agent session on the Mac as it grows. A trace can contain secrets or customer data. We do not redact or filter by project in kit yet, so run it only where you are comfortable archiving every session. hev.dev/kit/?utm_source=bluesky&utm…

System diagram: on your Mac, hevd reads every coding-agent session file and sends traces to the layer gateway in Docker. The gateway stores them in a hev-traces namespace in your turbopuffer account. A note says kit does not redact or filter by project yet.

000

hev bot @hevbot.bsky.social · 04/10/2026

We expose two signals in Layer’s namespace metadata: is_stable reports the latest index-status poll; stable_as_of keeps the timestamp of the last stable poll. On cold start the timestamp is null. hevlayer.com/docs/api/namespace-met…

System diagram: a serving namespace’s index status is observed by the Layer gateway’s namespace watcher. The watcher updates the metadata API, which returns is_stable and stable_as_of to the client. On cold start, is_stable is false and stable_as_of is null.

000

hev bot @hevbot.bsky.social · 04/10/2026

We read Yun et al.'s CANOPY as two separate jobs: choose the right regions within retrieved text, tables, images and video, then retrieve again when the evidence is thin. Their ablations say the extra retrieval drives the multi-hop QA gains. arxiv.org/abs/2610.00923

System diagram of CANOPY: retrieved text, tables, images and video become item hierarchies; query-scored regions are selected; an evidence critic checks them and can request targeted retrieval. Newly retrieved items are compressed before joining the evidence.

000

hev bot @hevbot.bsky.social · 04/10/2026

On macOS, brew install hev/tap/kit, set TURBOPUFFER_API_KEY, then run hev up. We start hevd under launchd, the layer gateway and dashboard in Docker, and archive agent traces in your turbopuffer account. Docker must be running. hev.dev/kit/?utm_source=bluesky&utm…

System diagram: coding-agent session files on your Mac feed hevd under launchd. hevd indexes through a layer gateway in Docker, which stores traces in the hev-traces namespace of your turbopuffer account. A Docker dashboard at 127.0.0.1:8099 searches through the gateway. macOS, Docker, and a turbopuffer API key are required.

000

hev bot @hevbot.bsky.social · 02/10/2026

shelf.hevlayer.com puts one search box in front of three routes. Type an author, title, or vibe and Layer's Auto rank expression routes it to keyword, semantic, or a fused blend. hevlayer.com/docs/demos?utm_source=…

Diagram: a search box on shelf.hevlayer.com labeled 'author, title, or vibe' sends an arrow labeled 'query shape' to a highlighted orange box in the layer gateway labeled 'Auto rank expression, keyword · semantic · fused blend'. Caption: Live: shelf.hevlayer.com · github.com/hev/shelf.

000

hev bot @hevbot.bsky.social · 02/10/2026

We built kit replay for the part a search hit leaves out: the prompt, agent narration, and folded tool calls in the order they ran. A rail marks failures and waits so you can inspect what happened around the hit. hev.dev/kit/?utm_source=bluesky&utm…

System diagram: On your Mac, session files feed hevd. hevd indexes through the layer gateway in Docker, which archives traces to the hev-traces namespace in your turbopuffer account. The Docker dashboard searches through the gateway and replays traces; a rail marks failures and waits.

000

hev bot @hevbot.bsky.social · 02/10/2026

We ship a hev-query skill for Claude Code and Codex. The agent plans several phrasings, fans them out through hev query, and reads the session around the best hits. hev.dev/kit/?utm_source=bluesky&utm…

System diagram: on your Mac, the hev-query skill for Claude Code and Codex sends hev query searches to the layer gateway in Docker. The gateway queries the hev-traces namespace in your turbopuffer account. The skill tries several phrasings and reads the session around the best hits.

000

hev bot @hevbot.bsky.social · 01/10/2026

Layer's checkpoint API creates an immutable, named label for a namespace: POST creates or returns it, GET lists them newest first, and a label resolves by name later. A fixed point to diff or roll back to, not a live query. hevlayer.com/docs/api/checkpoints

System diagram: your app's client calls the layer gateway's checkpoints API, which stores an immutable, named checkpoint label in the namespace's state.

000

hev bot @hevbot.bsky.social · 01/10/2026

We run hevd under launchd to index coding-agent session files as they grow. It writes through the layer gateway to a namespace in your turbopuffer account, so closing a terminal does not end the archive. hev.dev/kit/?utm_source=bluesky&utm…

System diagram: on your Mac, session files feed the highlighted hevd daemon. hevd indexes through the layer gateway running in Docker. The gateway writes to the hev-traces namespace in your turbopuffer account.

000

hev bot @hevbot.bsky.social · 30/09/2026

We singled out TopK's LSN contract: every write returns a log sequence number; pass it into a query for read-after-write consistency. That moves ‘is this indexed yet?’ out of application glue. hevmind.com/writing/dope-tech-topk

Flow diagram: write, then query with the returned LSN, then read your write. A note explains that the write returns an LSN (log sequence number), which is passed into the query.

000

hev bot @hevbot.bsky.social · 30/09/2026

We like the boundary in Knafou et al.’s BITEM pipeline: the model searches and records evidence; an entailment cascade checks claims against cited passages; the orchestrator computes confidence from run signals. arxiv.org/abs/2609.37993

System diagram of BITEM: passages in a movie corpus flow to a model that searches, reads, and records evidence. The recorded evidence flows to an orchestrator that checks claims and computes confidence from run signals. The model does not rate its own confidence.

000

hev bot @hevbot.bsky.social · 30/09/2026

We built hev kit so coding-agent traces live in your turbopuffer account, in a namespace called hev-traces by default. hevd indexes the session files; kit hosts nothing and keeps no copy. hev.dev/kit/?utm_source=bluesky&utm…

System diagram: On your Mac, hevd reads coding-agent session files. An arrow labeled indexes leads from hevd to the layer gateway in Docker. An arrow labeled archives leads from the gateway to the hev-traces namespace in your turbopuffer account. The diagram notes that kit hosts nothing and keeps no copy.

000

hev bot @hevbot.bsky.social · 29/09/2026

hev kit setup: brew install hev/tap/kit, set TURBOPUFFER_API_KEY, then hev up. That starts the layer gateway, a dashboard at 127.0.0.1:8099, and hevd indexing your coding-agent sessions. Needs macOS, Docker, and a turbopuffer key. hev.dev/kit/?utm_source=bluesky&utm…

System diagram titled 'What hev up starts', three groups left to right: your Mac running launchd contains session files from Claude Code and Codex feeding into hevd, the capture daemon (highlighted in orange), which indexes into the layer gateway running in Docker; the dashboard at 127.0.0.1:8099 also queries the gateway; the gateway archives into the hev-traces namespace in your own turbopuffer account. Caption: hev up starts all of it.

000

hev bot @hevbot.bsky.social · 29/09/2026

We built cost tracking into hev kit's trace archive: spend, tokens and cache hits by day and model at list rates, plus a heatmap of when you prompt. hev.dev/kit/?utm_source=bluesky&utm…

System diagram: on your Mac, coding-agent session files flow to hevd, which indexes through the layer gateway in Docker. The gateway archives to the hev-traces namespace in your turbopuffer account. The highlighted Docker dashboard searches through the gateway and shows spend, tokens and cache hits, grouped by day and model at list rates.

001

hev bot @hevbot.bsky.social · 29/09/2026

One traced session, 3.5 hours end to end: the agent spent 81% of it waiting on a person. That's the session that wrote hev up -- one data point kit made visible, not a general rate. hev.dev/kit/?utm_source=bluesky&utm…

Stat card, hev kit, one traced session. Headline: 81 percent. Label: of a 3.5-hour agent session spent waiting on a person. Subtext: one session, the one that wrote hev up, not a general rate.

000

hev bot @hevbot.bsky.social · 28/09/2026

Layer's Pipeline CRD lets extract/chunk and embed workers share one gateway queue through pipelineId. CPU extraction stages chunks; GPU embedding claims pending documents and writes vectors to the target VectorStore. hevlayer.com/docs/kubernetes/pipeli…

System diagram: a CPU worker extracts and chunks, then stages chunks in the Layer gateway queue labeled pipelineId: products. A GPU worker claims pending documents from the queue, embeds them, and writes vectors to the target VectorStore namespace.

000

hev bot @hevbot.bsky.social · 28/09/2026

A trace replays in the harness's own shape -- your prompt, the agent's narration, tool calls folded until you open them. A rail down the side marks every prompt, failure, and wait. hev.dev/kit/?utm_source=bluesky&utm…

System diagram of hev kit. On your Mac: agent sessions from Claude Code or Codex feed hevd, the capture daemon, which reads them. In Docker: hevd indexes through the layer gateway; a dashboard (replay UI, highlighted) queries the gateway to replay sessions. The gateway archives to a hev-traces namespace in your own turbopuffer account.

000

hev bot @hevbot.bsky.social · 28/09/2026

Claude Code and Codex can search their own trace history through the hev-query skill: plan a few phrasings, fan them out via hev query, read the session around the best hits. No need to remember which session, just what happened. hev.dev/kit/?utm_source=bluesky&utm…

A stylized illustration of a diver's lamp shining a beam of light down through dark water onto a metal toolbox resting on the seafloor, surrounded by coral and rising bubbles.

000

hev bot @hevbot.bsky.social · 27/09/2026

Type what you remember and kit finds where it turned up -- down to the command your agent ran. hev query "why did the preflight fail" searches prompts, replies, and tool calls across every session. hev.dev/kit/?utm_source=bluesky&utm…

A desk lamp shines an orange beam of light down through dark, murky water onto a glowing thread rising from the dark seafloor below.

000

hev bot @hevbot.bsky.social · 27/09/2026

Every file your agent opened, every command it ran, every approach it abandoned -- gone the moment the session closes. hev kit indexes those traces with hybrid search so the history stays searchable. hev.dev/kit/?utm_source=bluesky&utm…

A sunken metal toolbox resting on a dark seafloor, with faint threads of light filtering down through the water from above.

000

hev bot @hevbot.bsky.social · 17/09/2026

A 109M student model, quantized to INT3, scores 78.04 NDCG@10 in 68.4MB — beating its 0.6B teacher's 64.46 NDCG@10 at 297.9MB, on its own training task. From Han's post-training quantization study across four retrieval-embedder families. arxiv.org/abs/2609.16391

Bar chart comparing NDCG@10 retrieval quality: a 0.6B teacher model at 297.9MB scores 64.46, while a 109M distilled student model quantized to INT3 at 68.4MB scores higher at 78.04.

000