hypernix 0.72.5.dev3__tar.gz → 0.72.5.post3__tar.gz

{hypernix-0.72.5.dev3 → hypernix-0.72.5.post3}/.github/workflows/ios.yml

@@ -49,7 +49,12 @@ on:
       local_engine:
         description: "Build llama.cpp into the app so it can run models on-device."
         type: boolean
-        default: false
+        # True for a *called* build, unlike the dispatch default below.
+        # The only reason to call this workflow is to ship the result,
+        # and an IPA that cannot run a model is not worth shipping. The
+        # expensive default is the safe one here; the cheap default
+        # belongs on the PR path, where it is.
+        default: true
       version:
         description: "Marketing version to stamp into the build"
         type: string
@@ -131,8 +136,50 @@ jobs:
       # macOS runner and is not worth paying on a PR that touched a
       # view. Without it the app builds and ships with EchoRunner,
       # which tells the user this build has no local engine.
+      # Every trigger gets a stated answer, because the obvious
+      # expression does not have one.
+      #
+      # `if: ${{ inputs.local_engine }}` was the whole engine-skipping
+      # bug, and it looks correct. The `inputs` context exists *only*
+      # for `workflow_dispatch` and `workflow_call`; on a `push` or a
+      # `pull_request` it is not populated at all, so
+      # `inputs.local_engine` is null, null is falsy, and the step was
+      # skipped on every single commit to main -- silently, with a grey
+      # "skipped" in the log that looks exactly like a deliberate one.
+      # `--require-engine` was gated on the same expression, so the
+      # guard that exists to catch this was switched off by it too.
+      #
+      # Every commit to main therefore shipped an app whose on-device
+      # inference was EchoRunner. That is the "it always skips the
+      # inference engine build" report, and setting the workflow_call
+      # default to true (which fixed the release path) could not have
+      # fixed it: the release path was never the one skipping.
+      - name: Decide whether to build the engine
+        id: engine
+        env:
+          REQUESTED: ${{ inputs.local_engine }}
+          EVENT: ${{ github.event_name }}
+        run: |
+          set -euo pipefail
+          # REQUESTED is "true", "false", or empty when there is no
+          # `inputs` context to read. Empty is the case that has to be
+          # decided here rather than defaulted to false by accident.
+          case "$REQUESTED" in
+            true)  want=true ;;
+            false) want=false ;;
+            *)
+              # No input: the trigger decides. main is what people
+              # install from and it gets a real engine; a pull request
+              # does not, because 15-25 minutes on every push is what
+              # makes somebody turn this off permanently.
+              if [ "$EVENT" = "push" ]; then want=true; else want=false; fi
+              ;;
+          esac
+          echo "Engine requested: '${REQUESTED:-<no inputs context>}' on $EVENT -> $want"
+          echo "build=$want" >> "$GITHUB_OUTPUT"
+
       - name: Build the inference engine
-        if: ${{ inputs.local_engine }}
+        if: steps.engine.outputs.build == 'true'
         working-directory: ios
         run: |
           set -euo pipefail
@@ -151,7 +198,11 @@ jobs:
           # XcodeGen cannot make that call: its `optional: true` means
           # weak linking, not "skip when absent", so a spec that always
           # names the framework fails to build on a checkout without it.
-          python3 scripts/prepare_project.py
+          # --require-engine when the build asked for one, so a failed
+          # or skipped engine step fails here instead of shipping an IPA
+          # in which every on-device feature quietly does nothing.
+          python3 scripts/prepare_project.py \
+            ${{ steps.engine.outputs.build == 'true' && '--require-engine' || '' }}
           xcodegen generate --spec project.generated.yml
           ls -la HyperLink.xcodeproj
 
@@ -186,6 +237,21 @@ jobs:
           fi
           echo "Testing on simulator $device"
           set -o pipefail
+          # The whole log to a file, and only a summary to the console.
+          #
+          # This used to be `| tail -60`, which discards exactly what is
+          # needed when the build fails: xcodebuild prints a short
+          # "Testing failed:" summary of *deduplicated* messages and
+          # then the per-file errors with their line numbers, and the
+          # tail keeps the wrong end. A compile failure would report six
+          # unique messages and one file:line, so fixing it took a round
+          # trip per error.
+          #
+          # Now every `error:` line is echoed first, in full, and the
+          # complete log is uploaded as an artifact.
+          # The redirect below needs the directory to exist first.
+          mkdir -p build
+          set +e
           xcodebuild test \
             -project HyperLink.xcodeproj \
             -scheme HyperLink \
@@ -193,7 +259,28 @@ jobs:
             -destination "id=$device" \
             -resultBundlePath build/TestResults.xcresult \
             CODE_SIGNING_ALLOWED=NO \
-            | tail -60
+            > build/xcodebuild.log 2>&1
+          status=$?
+          set -e
+
+          if [ "$status" -ne 0 ]; then
+            echo "::group::Every compiler error"
+            # Sorted and deduplicated: a batch build repeats the same
+            # error once per file that triggered it, and forty copies of
+            # one message buries the other five.
+            grep -E "(error|warning): " build/xcodebuild.log \
+              | sed 's|/Users/runner/work/HyperNix-pip/HyperNix-pip/||' \
+              | sort -u \
+              | head -200 || true
+            echo "::endgroup::"
+            echo "::group::The last 200 lines"
+            tail -200 build/xcodebuild.log
+            echo "::endgroup::"
+            echo "::error::xcodebuild failed — see the groups above, or the ios-build-log artifact"
+          else
+            tail -20 build/xcodebuild.log
+          fi
+          exit "$status"
 
       - name: Archive
         working-directory: ios
@@ -315,6 +402,15 @@ jobs:
           if-no-files-found: error
           retention-days: 90
 
+      - name: Upload the build log
+        if: always() && inputs.run_tests != false
+        uses: actions/upload-artifact@v4
+        with:
+          name: ios-build-log-${{ github.run_number }}
+          path: ios/build/xcodebuild.log
+          if-no-files-found: ignore
+          retention-days: 7
+
       - name: Upload the test results
         if: always() && inputs.run_tests != false
         uses: actions/upload-artifact@v4

{hypernix-0.72.5.dev3 → hypernix-0.72.5.post3}/.github/workflows/public-release.yml

@@ -184,6 +184,19 @@ jobs:
 
           init = Path("src/hypernix/__init__.py")
           init.write_text(re.sub(r'__version__ = "[^"]+"', f'__version__ = "{v}"', init.read_text(), count=1))
+
+          # The fourth place a version string lives. install-t1.sh bakes
+          # one as the `curl | bash` fallback, for the case where there
+          # is no checkout to read and no hypernix installed yet to ask
+          # -- and a release that bumped the other three without this one
+          # shipped an installer whose banner announced the previous
+          # version over a correct install. tests/test_install_script.py
+          # fails when they drift, which is how this came to be here.
+          installer = Path("install-t1.sh")
+          installer.write_text(
+              re.sub(r'^VERSION="[^"]+"$', f'VERSION="{v}"',
+                     installer.read_text(), count=1, flags=re.M)
+          )
           PY
           echo "--- pyproject.toml version line"
           grep '^version' pyproject.toml
@@ -191,6 +204,8 @@ jobs:
           grep '^version' setup.cfg
           echo "--- __init__.py version line"
           grep __version__ src/hypernix/__init__.py
+          echo "--- install-t1.sh version line"
+          grep '^VERSION=' install-t1.sh
 
       - name: Lint
         run: ruff check src/hypernix
@@ -328,15 +343,63 @@ jobs:
 
       - name: Commit version bump
         if: ${{ !inputs.dry_run }}
+        env:
+          BRANCH: ${{ github.ref_name }}
+          VERSION: ${{ inputs.version }}
         run: |
           git config user.name  "github-actions[bot]"
           git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
-          git add pyproject.toml setup.cfg src/hypernix/__init__.py wiki/Release-Timeline.md
+
+          # Every file `Bump versions in source` writes. install-t1.sh
+          # is the one that was missing: the bump edited it and this
+          # list did not stage it, so the change was written, discarded,
+          # and the installer drifted back to the previous version on the
+          # very next release -- the bug the bump step was added to fix.
+          # tests/test_install_script.py checks the two lists agree.
+          git add \
+            pyproject.toml \
+            setup.cfg \
+            src/hypernix/__init__.py \
+            install-t1.sh \
+            wiki/Release-Timeline.md
+
           if git diff --cached --quiet; then
-            echo "::notice::no version change to commit (already at v${{ inputs.version }})"
-          else
-            git commit -m "hypernix v${{ inputs.version }} [skip ci]"
-            git push origin HEAD:${{ github.ref_name }}
+            echo "::notice::no version change to commit (already at v$VERSION)"
+            exit 0
+          fi
+          git commit -m "hypernix v$VERSION [skip ci]"
+
+          # Rebase and retry rather than failing outright.
+          #
+          # Minutes pass between checkout and here -- a full test suite,
+          # a build, a twine check -- and anything that lands on the
+          # branch in that window makes a plain push fail with "fetch
+          # first". That failure is expensive and badly timed: the
+          # artifacts are built and the *tag* step has not run, so the
+          # release ends up published with no tag pointing at it.
+          #
+          # The commit is one commit touching version files, so rebasing
+          # it onto whatever arrived is exactly right. A conflict means
+          # somebody else bumped the version concurrently, which is not
+          # something to resolve automatically -- that stops, loudly.
+          pushed=""
+          for attempt in 1 2 3 4 5; do
+            if git push origin HEAD:"$BRANCH"; then
+              pushed=yes
+              break
+            fi
+            echo "::warning::push $attempt rejected; rebasing onto origin/$BRANCH"
+            git fetch origin "$BRANCH"
+            if ! git rebase "origin/$BRANCH"; then
+              git rebase --abort || true
+              echo "::error::the version bump conflicts with $BRANCH -- another release may be in flight"
+              exit 1
+            fi
+            sleep $((attempt * 3))
+          done
+          if [ -z "$pushed" ]; then
+            echo "::error::could not push the version bump after 5 attempts"
+            exit 1
           fi
 
       - name: Tag and push

{hypernix-0.72.5.dev3 → hypernix-0.72.5.post3}/.github/workflows/release.yml

@@ -50,6 +50,14 @@ jobs:
     uses: ./.github/workflows/ios.yml
     with:
       run_tests: true
+      # A release IPA has to be able to run a model. Without this the
+      # build links no llama.cpp and ships EchoRunner, so every
+      # on-device feature is inert in the artifact people install --
+      # and nothing said so, because falling back is the correct
+      # behaviour for a PR build and the release used the same default.
+      # It costs 15-25 minutes and it is the difference between an app
+      # and a demo.
+      local_engine: true
       version: ""
     secrets: inherit
 

{hypernix-0.72.5.dev3 → hypernix-0.72.5.post3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hypernix
-Version: 0.72.5.dev3
+Version: 0.72.5.post3
 Summary: Download and quantize the HyperNix PyTorch model to GGUF (fp32/fp16/Q8_0/Q6_K/Q4_K_M). and train models with a massive ai toolkit 
 Home-page: https://github.com/trail-b1az3r/hypernix-pip
 Author: HyperNix contributors
@@ -78,7 +78,7 @@ Dynamic: license-file
 </p>
 
 <p align="center">
-  <img alt="PyPI" src="https://img.shields.io/badge/PyPI-v0.72.4.post20-ff2d55?style=for-the-badge&logo=pypi&logoColor=white" />
+  <img alt="PyPI" src="https://img.shields.io/badge/PyPI-v0.72.5.post1-ff2d55?style=for-the-badge&logo=pypi&logoColor=white" />
   <img alt="Python" src="https://img.shields.io/badge/Python-3.10--3.14-00c9ff?style=for-the-badge&logo=python&logoColor=white" />
   <img alt="License" src="https://img.shields.io/badge/License-HOS%20/%20LLU-00c853?style=for-the-badge" />
 </p>
@@ -87,12 +87,15 @@ Dynamic: license-file
 **End-to-end toolkit for training ai models on modern or old devices, originaly for converting hypernix.1 into gguf, now for all around training**
 
 ## What's fixed in this update
-See [Changelog.md](/wiki/Changelog.md)
-for most updates
+
+The section below covers what 0.72.5 adds.
+[`wiki/Changelog.md`](wiki/Changelog.md) is the canonical per-release
+history, and [`wiki/Roadmap.md`](wiki/Roadmap.md) says what is next.
 
 ## Table of contents
 
 - [What's fixed in this update](#whats-fixed-in-this-update)
+- [What's new: 0.72.5 — drafts, bundles, accounts, and a car](#whats-new-0725--drafts-bundles-accounts-and-a-car)
 - [What's new: 0.72.3.post2 — sub-bit models you can actually run](#whats-new-0723post2--sub-bit-models-you-can-actually-run)
 - [What's new: 0.72.3 — T1 v1.0.2026.8.1.1](#whats-new-0723--t1-v102026811)
 - [Package layout](#package-layout)
@@ -116,6 +119,150 @@ for most updates
 
 Cross-platform: Linux, macOS, Windows. Python 3.10 - 3.14.
 
+## What's new: 0.72.5 — drafts, bundles, accounts, and a car
+
+**Speculative decoding, in both shapes.** `hyprslug` produced a
+quantised model and `dflash2` could attach a draft somebody else had
+made. Now it makes the draft:
+
+```bash
+dflash2 attach model.gguf -o model+draft.gguf --quant q4_m   # embedded, one file
+dflash1        model.gguf -o draft.gguf       --quant int8   # for --model-draft
+```
+
+`dflash2` writes the draft into the base GGUF under a `dflash2.` prefix,
+which is where the speed-up is: one file, one download, and a runtime
+that has never heard of Dflash2 reads the base model straight through.
+`dflash1` writes it as its own file for `llama-cli --model-draft`,
+renumbering the kept blocks from zero and rewriting `<arch>.block_count`
+to match — a draft that claims 32 blocks and ships 6 loads, and then
+reads past the end of its own tensor table. It refuses a base with no
+tokenizer rather than writing a draft nothing can sample from.
+
+Both take nine targets: `q8`, `int8`, `fp16`, `bf16`, `fp32`, `IQ0.5`,
+`Q6_K`, `Q4_M`, `int2` — and anything else `hyprslug` writes, because
+they plan through it rather than keeping their own list. A draft at a
+given tier is the same packing as a model at it.
+
+**Several quantisations in one GGUF.**
+
+```bash
+hnx-bundle build model.gguf q8,q4_m,iq0.5 -o all.gguf
+hnx-bundle list all.gguf
+hnx-bundle extract all.gguf q4_m -o just-q4m.gguf
+```
+
+Variants live under `hnxq.<slug>.`, and the default one keeps the
+ordinary tensor names — a stock llama.cpp opens the file and runs it
+with nothing unusual happening. A bundle is roughly its variants added
+together: what is stored once is the tensors every variant left
+untouched byte for byte, which for most tiers is the 1-D norms rather
+than the bulk. The point is not compression — it is one download
+instead of five, one page cache instead of five, and switching tiers on
+a served model without fetching anything.
+
+**Five new quant targets.** `INT8` and `INT2` join the fixed-codebook
+family; `FP32` / `FP16` / `BF16` are element widths and take their own
+path. BF16 rounds to nearest even. FP16 saturates an overflow to 65504
+only when the input was *finite*, so an infinity stays an infinity
+instead of quietly becoming a large number that looks like data.
+
+INT2 was expected to lose to FP2 and does not — 0.384 relRMS against
+0.397 over twelve seeds, because ~40% of a weight tensor sits near zero
+and INT2 spends a codeword there. Measurement in
+[LowBit](wiki/LowBit.md).
+
+**`tvtoppro` gets an intro, modules, and a watchdog.** `tvtop-older`'s
+animated "decoding" startup text and the spinner, in tvtoppro's
+presentation; a module system, so a new stat is a file that registers
+itself rather than a patch to the renderer; and a stall detector —
+`train.log` untouched for over a week means the run being watched is not
+the run that is happening, so it finds the busiest Python process on the
+machine and reads *its* logs and progress instead.
+
+```bash
+tvtoppro --find-run          # go looking, rather than waiting on a dead log
+```
+
+**`cctvtop`'s Remote Desktop panel works.** It called a session live
+whenever *something* held the port. The probe completes an RFB handshake
+now, so a stale listener, a tunnel with nothing behind it and a real
+desktop are three different answers — and when there is no session it
+says which of four reasons applies rather than "unavailable".
+
+**`noodle` runs inside `hyped-pro`.** `/noodle` in the TUI. It adopts
+vendor keys already stored in the HyperNix config, with the environment
+always winning over the stored copy, so a session that works in `hyped`
+is not configured twice.
+
+**The `hypernix` CLI says what was wrong.** An unknown subcommand used to
+print usage and exit 0. It now goes to stderr, exits 2, and suggests the
+nearest real command:
+
+```
+$ hypernix quantise
+hypernix: unknown command 'quantise'
+
+  You want:  hypernix quantize
+
+  hypernix --help   every subcommand
+```
+
+**T1 v1.0.26.9.2.3 — an account without already having a key.** Sign-up
+and browser sign-in, served four ways: from localhost, over Tailscale,
+from the operator's own site, or from a prebuilt Cloudflare site hosted
+by the API host.
+
+```bash
+t1-accounts create mason --admin     # password prompted, never an argument
+t1-accounts modes                    # what the four are, and what each needs
+
+T1_ACCOUNTS_ENABLED=1 T1_ACCOUNTS_MODE=tailscale hypernix-t1 start
+```
+
+scrypt for passwords, constant-time comparison, CSRF on every form,
+SameSite cookies, lockout after repeated failures, and `Secure` coupled
+to whether the connection is actually TLS. A keyless caller never gets
+administrator rights, in any of the four modes.
+
+**Two security fixes.** The config file holding API keys was
+world-readable; it is now written `0600` into a `0700` directory through
+a temporary file and `os.replace`, and an existing file is tightened on
+load. And key authentication compared key *strings*, in a loop whose
+position depended on the prefix — a timing oracle for the stored keys.
+Keys are indexed by digest and compared with `hmac.compare_digest`; the
+measured 60x spread across probe keys is flat.
+
+**HyperLink in a car, and out loud.** The iOS app gets a CarPlay scene,
+four Siri intents, eight themes, an attachment menu that offers all four
+ways in, and renaming.
+
+- CarPlay shows a conversation list, `CPVoiceControlTemplate` dictation
+  and six canned replies — and a keyboard **only when the car reports it
+  will allow one**, read from `CPSessionConfiguration.limitedUserInterfaces`
+  on every use and rebuilt from its delegate, because the answer changes
+  while the app is running.
+- "Siri, ask HyperLink what's the weather", "Siri, load the model Gemma 4
+  E2B on blazeindustries in HyperLink", "Siri, read me the HyperLink
+  chat". None of the four intents opens the app; the point of asking from
+  a car dock is that the phone stays where it is.
+- Every theme's text clears WCAG AA against its own bubble, the two
+  bubbles are told apart by luminance, and so are "connected" and
+  "failed" — checked in `tests/test_hyperlink_ios_wiring.py`, which found
+  three that were not.
+- A chat can be renamed. The server has taken a title on
+  `PATCH /hyperlink/sessions/{id}` since HyperLink shipped, and nothing
+  on the phone ever sent one.
+
+CarPlay needs `com.apple.developer.carplay-communication`, which Apple
+grants per app on request rather than with a paid account. Everything
+else builds and runs without it.
+
+**Docs.** Issue templates for bugs and for feature requests, and
+[Model-Training-Guide](wiki/Model-Training-Guide.md) — which of these do
+I use, and in what order. Every API in that guide was run against the
+package rather than written from memory; six were wrong.
+
 ## What's new: 0.72.3.post2 — sub-bit models you can actually run
 
 **The error this release is about.** LM Studio, opening a model this
@@ -237,21 +384,26 @@ directory:
 
 | Directory | Modules | Contents |
 |---|---|---|
+| `hypernix/audio/` | 4 | Wake words, audio files and speech plumbing. |
 | `hypernix/chat/` | 5 | Chat templating, prompt presets and multi-turn session state. |
-| `hypernix/data/` | 15 | Datasets: collection, cleaning, splitting, packing and augmentation. |
+| `hypernix/data/` | 17 | Datasets: collection, cleaning, splitting, packing and augmentation. |
 | `hypernix/evaluation/` | 6 | Scoring, rubric labelling, judging and module verification. |
-| `hypernix/interfaces/` | 11 | Human-facing front ends: CLIs, TUIs, GUIs and launchers. |
-| `hypernix/models/` | 11 | Architectures, snapshot loading, generation and model utilities. |
-| `hypernix/monitoring/` | 9 | Live dashboards, logging, telemetry and hardware sampling. |
-| `hypernix/optimizers/` | 8 | The Pressure Cooker optimizer family and optimizer plumbing. |
-| `hypernix/quant/` | 4 | The GGUF pipeline: convert, quantize, fetch tooling and upload. |
-| `hypernix/security/` | 3 | API keys, quotas and request gating. |
-| `hypernix/system/` | 15 | Environment, dependencies, hardware and housekeeping. |
+| `hypernix/hyperlink/` | 12 | The [HyperLink](wiki/HyperLink-Sync.md) server side: sessions, attachments, pairing, push. |
+| `hypernix/interfaces/` | 19 | Human-facing front ends: CLIs, TUIs, GUIs, launchers and `noodle`. |
+| `hypernix/models/` | 17 | Architectures, snapshot loading, generation and model utilities. |
+| `hypernix/monitoring/` | 14 | Live dashboards, logging, telemetry and hardware sampling. |
+| `hypernix/optimizers/` | 9 | The Pressure Cooker optimizer family and optimizer plumbing. |
+| `hypernix/quant/` | 28 | The GGUF pipeline: convert, quantise, the sub-bit tiers, the draft builders, bundles, fetch tooling and upload. |
+| `hypernix/scriptgen/` | 6 | The training-script builder behind `hnx-scriptgen`. |
+| `hypernix/security/` | 6 | API keys, quotas and request gating. |
+| `hypernix/system/` | 24 | Environment, dependencies, hardware and housekeeping. |
 | `hypernix/timing/` | 5 | Timers, alarms, cadence control and progress animation. |
-| `hypernix/training/` | 14 | Training entry points, schedules and weight perturbation. |
-| `hypernix/t1api/` | — | The [T1 API](wiki/T1-API.md) server: registry, routing, quota, billing, audit, rate limiting, mTLS, deployment. |
-| `hypernix/t1sdk/` | — | The T1 API client SDK — typed, stdlib-only, no server extra needed. |
-| `hypernix/waiter/` | — | [`waiter`](wiki/Waiter-TUI.md), the official T1 API TUI/CLI. |
+| `hypernix/training/` | 15 | Training entry points, schedules and weight perturbation. |
+| `hypernix/t1api/` | 57 | The [T1 API](wiki/T1-API.md) server: registry, routing, quota, billing, audit, rate limiting, mTLS, accounts, web auth, deployment. |
+| `hypernix/t1sdk/` | 4 | The T1 API client SDK — typed, stdlib-only, no server extra needed. |
+| `hypernix/waiter/` | 7 | [`waiter`](wiki/Waiter-TUI.md), the official T1 API TUI/CLI. |
+
+Counts are `.py` files excluding `__init__.py`, as of 0.72.5.
 
 **Every module keeps its old import path.** `hypernix.timer` and
 `hypernix.timing.timer` return the same module object, so nothing that
@@ -413,12 +565,9 @@ Eleven major additions:
 - **tvtop++ fixes** — Eliminated border flicker (layout built once), added `_block_history_bar` re-export, implemented `small_mode`, fixed self-process filtering.
 - **New wiki pages** — [Pressure-Cooker-V5](wiki/Pressure-Cooker-V5.md), [MTP](wiki/MTP.md), [Scavenger](wiki/Scavenger.md)
 - **Kitchen.md updated** — Added scavenger, MTP, and QAT sections
-- **Training benefits chart** — See below
 
 ### Training Benefits vs Complexity
 
-![HyperNix Training Features](docs/public/training-benefits-chart.png)
-
 > **Key insight**: MTP + Speculative Decoding offer the highest benefit-to-cost ratio. Int8-quantized momentum cuts the momentum buffer's own memory by 75% versus fp32, and PressureCookerV5/V5S's factored curvature keeps the rest of the optimizer state small too -- measured optimizer-state memory lands around 12-13% of AdamW's (see the [efficiency paper](pressure_cooker_v5_v5s_paper.md) for the exact numbers and methodology). The trade-offs -- including step-time overhead on some hardware -- are real and are covered in the paper rather than summarized as a single percentage here.
 
 ## What's new in v0.70.4
@@ -599,6 +748,10 @@ new_fridge.plot_loss_curve(new_fridge.parse_training_log(log), "loss.png")
 
 ## CLI reference
 
+`hypernix --help` prints the authoritative list — this is the shape of
+it. An unknown subcommand exits 2 on stderr and suggests the nearest
+real one rather than printing usage and claiming success.
+
 ```
 hypernix <subcommand> [options]
 
@@ -618,11 +771,39 @@ hypernix <subcommand> [options]
   generate              sample text from a local snapshot
   oven                  code-generation wrapper (preheat + complete / fill)
   chat                  interactive chat REPL against any supported model
-  hyped+ / hyped-pro    Node.js TUI agent CLI w/ real cloud+local model dispatch, /gui desktop mode
-                        (/t1api routes through a local or remote HyperNix T1 API server)
+  brew                  custom architecture builder and training suite
+  fizzle                fuse models and LoRAs                        (fiz)
+  fusebox               GPU thermal governor: pace a run, trip on heat
+  devices               what accelerators this machine has, and what they do
+  wakeup                train a trigger phrase, then listen for it
+  websearch             the search backend behind chat and the assistant
+  gather                crawl a site into a corpus (robots-aware)
+  runtime               use HyperNix's llama.cpp from LM Studio and friends
+  hyprslug-headers      self-describing GGUF headers, and the sub-bit server
   stml                  VRAM trained context length calculator
 ```
 
+The rest are their own console scripts, because each is a program rather
+than a step in the pipeline:
+
+| Command | What it is |
+|---|---|
+| `hyped` / `hyped-pro` | The TUI agent CLI. `/gui` for a desktop window, `/t1api` to route through a T1 server, `/noodle` for the autonomous executor. |
+| `hyprslug` (`doomslug`, `dstd`) | The quantiser, including the sub-bit tiers. |
+| `dflash2` / `dflash1` | Speculative-decoding drafts: embedded in the base GGUF, or standalone for `--model-draft`. |
+| `hnx-bundle` (`multiquant`) | Several quantisations in one GGUF. |
+| `hnx-imatrix` | Importance matrices. |
+| `steamroller` | The descending llama.cpp quantiser. |
+| `tvtoppro` / `cctvtop` / `tvtop-old` / `tvtop-older` | The dashboards, newest to oldest. |
+| `hnx-map` | The steampunk schematic TUI. |
+| `hnx-scriptgen` | The training-script builder. |
+| `waiter` | The T1 API's client TUI/CLI. Needs no server extra. |
+| `t1-accounts` | Web accounts for a T1 server: create, reset, unlock, and the four deployment modes. |
+| `gkey` | Gatekeeper + Keymaster, in one place. |
+| `multilama` | One interface over vanilla llama.cpp, ik_llama.cpp, PrismML and KoboldCpp. |
+| `noodle` | The autonomous multi-agent executor, standalone. |
+| `eth` / `ups` | Ethanol, and the UPS monitor. |
+
 `train run` accepts curriculum / context management flags:
 
 ```bash
@@ -723,6 +904,15 @@ Topic-focused reference guides live in the `wiki/` directory:
 - [`wiki/Pressure-Cooker-V5.md`](wiki/Pressure-Cooker-V5.md) — Pressure Cooker V5 / V5+ / V5S with QAT and MTP
 - [`wiki/MTP.md`](wiki/MTP.md) — Multi-Token Prediction guide
 - [`wiki/Scavenger.md`](wiki/Scavenger.md) — HuggingFace dataset discovery
+- [`wiki/Model-Training-Guide.md`](wiki/Model-Training-Guide.md) — which of these do I use, and in what order
+- [`wiki/HyprSlug.md`](wiki/HyprSlug.md) — the sub-bit quantiser
+- [`wiki/HyprSlug-Headers.md`](wiki/HyprSlug-Headers.md) — self-describing headers, the compat export, the sub-bit server
+- [`wiki/LowBit.md`](wiki/LowBit.md) — the fixed-codebook tiers and the measurements behind them
+- [`wiki/Dflash2.md`](wiki/Dflash2.md) — speculative-decoding drafts
+- [`wiki/TvTopPro.md`](wiki/TvTopPro.md) — the dashboard, its themes and its modules
+- [`wiki/T1-API.md`](wiki/T1-API.md) — the T1 API server
+- [`wiki/HyperLink-Sync.md`](wiki/HyperLink-Sync.md) — the phone app and what it talks to
+- [`wiki/Roadmap.md`](wiki/Roadmap.md) — what is planned, and what landed
 
 ## How the GGUF pipeline works