Skip to content

Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.

C++

Docs: plain-wording pass on the 0.2.0 text (fewer mid-sentence bolds, direct phrasing)

Niko12215e7b9a83015e
diff --git a/AGENTS.md b/AGENTS.md
index f2696dfff..89efd187d 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -14,7 +14,7 @@ or an Apple Silicon Mac. Version 0.2.0-beta, engine 0.1.45.3. It has three parts
## Installing Strata for a user
-Follow **[docs/AI_SETUP.md](docs/AI_SETUP.md)**. The short version, for an agent that has a shell:
+Follow **[docs/AI_SETUP.md](docs/AI_SETUP.md)**. For an agent that has a shell, the short version is:
1. Install the Terminal. It needs no administrator rights.
diff --git a/SECURITY.md b/SECURITY.md
index b279eeebc..8b05515e2 100644
--- a/SECURITY.md
+++ b/SECURITY.md
@@ -83,12 +83,12 @@ What the script does is short and written at its top: it fetches the Terminal ar
checks its SHA-256, unpacks it into your home folder (no administrator rights, nothing outside your home), checks the release's signed feed
with the Terminal it just unpacked and removes it again if that fails, and puts `strata` on your PATH.
-Know the limit of those checks. The checksum comes from the same release as the archive, and the public key that checks the feed arrives inside
-that archive, so for the **first** install all trust rests on HTTPS and on the `Niko1221/Strata` GitHub release itself. The checks catch a
+Those checks have a limit. The checksum comes from the same release as the archive, and the public key that checks the feed arrives inside
+that archive, so for the first install all trust rests on HTTPS and on the `Niko1221/Strata` GitHub release itself. The checks catch a
damaged or half-finished download and a feed that was not signed by the release key; they do not protect you if the release has been replaced
by someone who controls it. Later updates are checked against the key already on your PC, which is a stronger position.
-If that is not enough for you: save the script (`curl -fsSLO https://github.com/Niko1221/Strata/releases/latest/download/install.sh`), read it, and
+If that is not enough: save the script (`curl -fsSLO https://github.com/Niko1221/Strata/releases/latest/download/install.sh`), read it, and
run it from the file; or install the desktop app or build from source instead and check the code you build.
Strata has not had an outside security audit yet.
diff --git a/docs/AI_SETUP.md b/docs/AI_SETUP.md
index d9a083d4d..3f8e9c988 100644
--- a/docs/AI_SETUP.md
+++ b/docs/AI_SETUP.md
@@ -94,8 +94,8 @@ strata launch --plan --model iq2_xs # what would happen; nothing is c
strata launch --model iq2_xs --yes --no-tui # do it: download the engine and the model, start the server
```
-With `--no-tui` and `--yes` it asks nothing, prints the addresses when the server is ready, and returns. **The server
-keeps running** in the background. `strata launch` with no `--model`, `--preset` or other choice needs a terminal for
+With `--no-tui` and `--yes` it asks nothing, prints the addresses when the server is ready, and returns. The server
+keeps running in the background. `strata launch` with no `--model`, `--preset` or other choice needs a terminal for
its screens, so an agent always names the model. Add `--output json` for output a script can read.
The flags (all of them, with examples: `strata launch --help`):
@@ -176,7 +176,7 @@ With an API key, add `-H "Authorization: Bearer <key>"`.
- **Browser:** `http://127.0.0.1:8080` - Chat, a live Monitor, and About (settings and addresses).
- **Any OpenAI-compatible app or agent:** base URL `http://127.0.0.1:8080/v1`, any API key (or the configured one),
any model name.
-- **Anthropic-compatible apps:** base URL `http://127.0.0.1:8080`, **without `/v1`**. Those apps add `/v1/messages`
+- **Anthropic-compatible apps:** base URL `http://127.0.0.1:8080`, without `/v1`. Those apps add `/v1/messages`
themselves.
- **Codex CLI and other Responses API apps:** `http://127.0.0.1:8080/v1/responses` (stateless; Codex's
`config.toml`: [DETAILS.md](DETAILS.md#the-responses-api-and-codex-cli)).
@@ -200,7 +200,7 @@ git clone https://github.com/Niko1221/Strata.git
```
Without git: download https://github.com/Niko1221/Strata/archive/refs/heads/main.zip and unzip it. Pick a drive with
-enough free space: the model files go to a `Strata-data` folder **next to** the Strata folder (or `--data-dir <path>`
+enough free space: the model files go to a `Strata-data` folder next to the Strata folder (or `--data-dir <path>`
on another drive). Setup's own check prints the GPU(s), driver, RAM, CPU, free disk and whether this PC can run
Strata, and installs nothing except Python and the `.venv` it needs to run:
@@ -230,7 +230,7 @@ The flags (all of them: `START-HERE.bat --help`):
| `--backend cuda\|hip` | NVIDIA or AMD engine; chosen by itself on a PC with only one kind of card |
| `--data-dir PATH` | where the 70-120 GB of model files go |
| `--port N` | the server port (default 8080) |
-| `--host H --api-key K` | listen beyond this PC; **only together with a key** |
+| `--host H --api-key K` | listen beyond this PC; only together with a key |
| `--no-start` | install only, do not start the server |
| `--setup` | install another model or change settings of an installed one |
| `--check` | only check the PC |
diff --git a/docs/INSTALL.md b/docs/INSTALL.md
index 58cfe8ecc..ec533147f 100644
--- a/docs/INSTALL.md
+++ b/docs/INSTALL.md
@@ -24,7 +24,7 @@ There are three ways to install it. Take the first one unless you have a reason
| **GPU** | **NVIDIA** RTX 20, 30, 40 or 50 series, **12 GB VRAM or more** (8 GB runs, slowly). Measured on an RTX 5070 and an RTX 3090; RTX 20 (Turing, since 0.1.27) was tested by a contributor on an RTX 2070. **AMD** Radeon RX 7900 XT / XTX, RX 9070 / 9070 XT and Radeon AI PRO R9700 (validated), RX 7800 XT / 7700 XT and RX 9060 XT (validated by their owners), RX 6800 / 6900 series (community-reported), with 12 GB of VRAM or more. See [AMD cards](#amd-cards). |
| **RAM** | Enough for the size you pick ([which model](MODELS.md#pick-by-ram)); **64 GB** runs every size. A big GPU makes up for less RAM - the [low-RAM mode](MODELS.md#a-big-graphics-card-and-little-ram). |
| **CPU** | x86-64 with AVX2 (any Intel/AMD desktop CPU from the last ~8 years). AVX-512 (Ryzen 7000/9000) is a bit faster. Older CPUs without AVX2 are experimental and slow: [Older CPUs](#older-cpus-experimental). |
-| **Disk** | The app and its engine take about 1-3 GB. ~70-80 GB free for the model, ~6 GB for the MTP layer (+1 GB with images). **Q2_0 on an AVX-512 CPU** also writes a one-time ~40 GB copy of its experts for the fast CPU kernel. On Linux with an AMD card, ROCm takes ~10 GB more when setup installs it. An NVMe SSD is strongly recommended: it makes the first start much faster. |
+| **Disk** | The app and its engine take about 1-3 GB. ~70-80 GB free for the model, ~6 GB for the MTP layer (+1 GB with images). Q2_0 on an AVX-512 CPU also writes a one-time ~40 GB copy of its experts for the fast CPU kernel. On Linux with an AMD card, ROCm takes ~10 GB more when setup installs it. An NVMe SSD is strongly recommended: it makes the first start much faster. |
| **OS** | Windows 10/11, Linux (Ubuntu 22.04/24.04 and Debian 12/13 get everything installed automatically; the app also runs on Fedora 44), or a Mac with Apple Silicon (M1 or newer) and macOS 15 or newer ([Mac](MAC.md)). |
| **Driver** | **NVIDIA:** a current driver, version 580 or newer ([nvidia.com/drivers](https://www.nvidia.com/drivers) or the NVIDIA App). **AMD:** on Linux the kernel's amdgpu driver (no ROCm install needed); on Windows a current AMD Software: Adrenalin Edition driver ([amd.com/support](https://www.amd.com/en/support)). |
diff --git a/docs/MAC.md b/docs/MAC.md
index 59ff725df..ce9c81116 100644
--- a/docs/MAC.md
+++ b/docs/MAC.md
@@ -190,8 +190,8 @@ The arena was 11.5-14.7 GB. Tokens per second, one token per step (no MTP), medi
- **Images (vision): the source build only.** `./setup.sh` asks "Should the model also read images?" (`--vision yes|no`) and
builds the picture reader (`strata-vision`) itself. Checked on one test picture with Q2_0 and UD-Q4_K_XL: the answers equal
- the HIP engine's, and text output is unchanged with vision on. No other tests yet. **The app and the Terminal have no
- picture reader on the Mac yet**: they show models as text only and do not turn pictures on.
+ the HIP engine's, and text output is unchanged with vision on. No other tests yet. The app and the Terminal have no
+ picture reader on the Mac yet. They show models as text only and do not turn pictures on.
- **Context:** measured up to 32K (needles, KL). Setup recommends 32K below 96 GB and 64K from 96 GB. Up to the model's 262K can be set, but it is not measured.
- **One request at a time.** `--parallel` (batch slots) is not ported. Several clients queue.
- **Not ported:** the control vector (`--control-vector-scaled`) and the PC tuning (`--calibrate`).
diff --git a/docs/MODELS.md b/docs/MODELS.md
index fee706341..601bbee23 100644
--- a/docs/MODELS.md
+++ b/docs/MODELS.md
@@ -94,13 +94,13 @@ Coder. With a small card most experts then come from the SSD and it is much slow
## The versions
-**Pictures.** The versions in setup's menu read pictures, and pictures are **on by default** wherever a model can
+**Pictures.** The versions in setup's menu read pictures, and pictures are on by default wherever a model can
read them. The picture reader runs on the CPU when the graphics card has less than 24 GB (a few seconds per
picture, nothing taken from the card) and on the graphics card from 24 GB (about 1-1.4 GB of VRAM; 0.1-0.5 s per
picture). The rule is the same on NVIDIA and AMD cards. `--vision off` (or "No" in the app and in `strata launch`) turns
pictures off; `--vision cpu` or `--vision gpu` picks the reader. On AMD, `strata-vision-hip` is the card's reader; if it
-cannot start, Strata uses the CPU reader by itself ([AMD_HIP.md](AMD_HIP.md)). **The Mac app and the Mac Terminal have no
-picture reader yet** ([MAC.md](MAC.md)), and neither does the Intel Arc engine ([INTEL.md](INTEL.md#not-done)). More: [Images](DETAILS.md#images-vision).
+cannot start, Strata uses the CPU reader by itself ([AMD_HIP.md](AMD_HIP.md)). The Mac app and the Mac Terminal have no
+picture reader yet ([MAC.md](MAC.md)), and neither does the Intel Arc engine ([INTEL.md](INTEL.md#not-done)). More: [Images](DETAILS.md#images-vision).
### Qwen3.8-Flash-Next
@@ -173,7 +173,7 @@ experts in a rotated basis. They are the smallest options and never setup's defa
| **Gyro-S** | 58.5 GB | 27.6 GB | the smallest option, for PCs with little memory; lower quality |
| **Gyro-M** | 92.0 GB | 35 GB | about IQ3_XXS quality with about 20% less GPU memory; decodes about 20% slower than Q2_0 on 12 GB cards |
-NVIDIA, and AMD (experimental); not Intel Arc or the Mac. Gyro **reads pictures** (the same rule as above). It needs
+NVIDIA, and AMD (experimental); not Intel Arc or the Mac. Gyro reads pictures (the same rule as above). It needs
engine 0.1.45.1 or newer; if the engine is older, the app says "This model needs a newer engine" and offers Update
engine. Gyro gives correct output on RTX 50 cards with engine 0.1.45.3. Details, measurements and credit:
[GYRO.md](GYRO.md).
@@ -195,7 +195,7 @@ versions of it. The app shows them by who made the weights:
| **NVFP4 (NVIDIA's version)** | `NVFP4-NVIDIA` | NVIDIA's own NVFP4 checkpoint (plain rounding), converted for Strata. The same size and speed as the first | 128 GB |
| **NVFP4 (NVIDIA, trained for FP4)** | `NVFP4-QAD` | NVIDIA's retrained 4-bit model, slightly faster on short prompts ([below](#nvfp4-qad)) | 102 GB (+110 GB of disk to prepare) |
-All three need about 80 GB of RAM to keep every expert in RAM, run on **NVIDIA cards only**, and read pictures (the
+All three need about 80 GB of RAM to keep every expert in RAM, run on NVIDIA cards only, and read pictures (the
same rule as above). RTX 50 cards have FP4 tensor cores; other NVIDIA cards run NVFP4 at 8-bit precision. The first two
need engine 0.1.44 or newer, the third 0.1.45 or newer. Details and measurements: [NVFP4.md](NVFP4.md).
diff --git a/docs/QWEN36.md b/docs/QWEN36.md
index 3f3ecd36d..a2455b827 100644
--- a/docs/QWEN36.md
+++ b/docs/QWEN36.md
@@ -13,7 +13,7 @@ UD-Q4_K_XL in the hybrid mode: 22-37 of 256 experts per layer in VRAM, the rest
The vision encoder is Unsloth's `mmproj-F16.gguf` (0.9 GB, pinned revision, SHA-256 checked by setup): a 27-layer ViT with a 2x2 merger,
the same projector family as the Flash-Next models' (`qwen3vl_merger`; only its output width differs: 2048 here, 2560 there). Strata runs
it with the same helper (`strata-vision`, llama.cpp's `mtmd`), so the encoder choices are the Flash-Next ones
-([Images](DETAILS.md#images-vision)). **Pictures are on by default**, the same rule as for every model on NVIDIA and AMD cards:
+([Images](DETAILS.md#images-vision)). Pictures are on by default, the same rule as for every model on NVIDIA and AMD cards:
the CPU reader under 24 GB of graphics memory (a few seconds per picture, no VRAM), the card's reader from 24 GB (GPU encoder,
~1.2 GB of VRAM at up to 1,024 image tokens; on AMD it is `strata-vision-hip`, with the CPU reader as its fallback).
`--vision yes|no|gpu|cpu` and `--vision-tokens N` work as there; what you ask for is kept, with a note. The Mac engine does
diff --git a/docs/TERMINAL.md b/docs/TERMINAL.md
index 40c6922bc..0165615a9 100644
--- a/docs/TERMINAL.md
+++ b/docs/TERMINAL.md
@@ -91,7 +91,7 @@ strata runtime restart --profile four-gpu --allow-source-migration --plan --outp
strata runtime restart --profile four-gpu --allow-source-migration --if-profile-revision PROFILE_REVISION --if-source-sha256 CANDIDATE_SOURCE_SHA256
```
-The local host owns its server child. On macOS and Linux it uses a private Unix control socket and token file; on Windows it uses a loopback-only TCP listener (127.0.0.1, never a network address) that answers only requests carrying the owner's 256-bit token, kept in an owner-only folder (that is a weaker guarantee than a socket the other users cannot open, and the token is what protects it); the CLI never signals a PID read from a ready file. Restart drains the current instance, applies the saved profile revision, waits for the new ready file, and launches the previous arguments again if activation fails. Switching the saved profile to a **distinct, regular source file** is explicit and requires the preview's source digest. The old source file must still contain its launch bytes for ordinary rollback. An in-place edit requires `--allow-source-migration`; before drain the host copies the running config bytes to a mode-0600 file in its private runtime directory. If the old file was edited **and** the saved profile now points to a new file, preview and run with both source flags. A failed activation restarts from the snapshot, leaving the operator's edited file untouched. The rollback status reports the snapshot path; after the operator fixes the candidate file, retry it as a distinct source with `--allow-source-switch`. A successful activation removes the unneeded snapshot. A hard link to the old file or a source change after preview is rejected before draining; a source change during activation fails attestation and attempts rollback. The host stores private logs and state under `${XDG_STATE_HOME:-~/.local/state}/strata/runtime` or `STRATA_RUNTIME_DIR` (which must be an absolute, owner-only directory). `runtime start` and `runtime restart` verify the ready-file instance ID against authenticated local telemetry, compare a private MAC of the config bytes and launch arguments actually read by the server, and run a one-token generation before reporting generation readiness. A launch mismatch fails before generation. The MAC is scoped to the owner-managed process; it is not a publisher signature or a model-file hash. An intentionally lazy model remains `unloaded` with `generation_ready=false`; `runtime verify --load` opts into loading and generating. `runtime status` reports the last successful startup or explicit verification, not continuous engine liveness; use `runtime verify` to recheck. The probe validates the bound endpoint and model ID, requires at least one completion token, uses a configured API key only in memory, and puts neither generated text nor credentials in the operation journal. If the optional API monitor is enabled, it can capture the fixed probe request like other inference calls. This one-token smoke check does not establish sustained throughput, answer quality or four-card stability. On macOS and Linux, an inherited pipe makes the server stop its dedicated process group when the owner host dies, including after a hard kill. `runtime status` then reports `stale`; a new `runtime start` waits for the recorded child to exit, verifies the host's file lease is gone and the child has exited, and reclaims private state and any orphaned rollback snapshots without signaling a saved PID. If the child is still present, its ownership record is missing, or the socket has an active listener, recovery refuses and requires inspection. The desktop app starts its models through this host on all three systems. Multi-GPU workloads have not been qualified on this path.
+The local host owns its server child. On macOS and Linux it uses a private Unix control socket and token file; on Windows it uses a loopback-only TCP listener (127.0.0.1, never a network address) that answers only requests carrying the owner's 256-bit token, kept in an owner-only folder (that is a weaker guarantee than a socket the other users cannot open, and the token is what protects it); the CLI never signals a PID read from a ready file. Restart drains the current instance, applies the saved profile revision, waits for the new ready file, and launches the previous arguments again if activation fails. Switching the saved profile to a distinct, regular source file is explicit and requires the preview's source digest. The old source file must still contain its launch bytes for ordinary rollback. An in-place edit requires `--allow-source-migration`; before drain the host copies the running config bytes to a mode-0600 file in its private runtime directory. If the old file was edited and the saved profile now points to a new file, preview and run with both source flags. A failed activation restarts from the snapshot, leaving the operator's edited file untouched. The rollback status reports the snapshot path; after the operator fixes the candidate file, retry it as a distinct source with `--allow-source-switch`. A successful activation removes the unneeded snapshot. A hard link to the old file or a source change after preview is rejected before draining; a source change during activation fails attestation and attempts rollback. The host stores private logs and state under `${XDG_STATE_HOME:-~/.local/state}/strata/runtime` or `STRATA_RUNTIME_DIR` (which must be an absolute, owner-only directory). `runtime start` and `runtime restart` verify the ready-file instance ID against authenticated local telemetry, compare a private MAC of the config bytes and launch arguments actually read by the server, and run a one-token generation before reporting generation readiness. A launch mismatch fails before generation. The MAC is scoped to the owner-managed process; it is not a publisher signature or a model-file hash. An intentionally lazy model remains `unloaded` with `generation_ready=false`; `runtime verify --load` opts into loading and generating. `runtime status` reports the last successful startup or explicit verification, not continuous engine liveness; use `runtime verify` to recheck. The probe validates the bound endpoint and model ID, requires at least one completion token, uses a configured API key only in memory, and puts neither generated text nor credentials in the operation journal. If the optional API monitor is enabled, it can capture the fixed probe request like other inference calls. This one-token smoke check does not establish sustained throughput, answer quality or four-card stability. On macOS and Linux, an inherited pipe makes the server stop its dedicated process group when the owner host dies, including after a hard kill. `runtime status` then reports `stale`; a new `runtime start` waits for the recorded child to exit, verifies the host's file lease is gone and the child has exited, and reclaims private state and any orphaned rollback snapshots without signaling a saved PID. If the child is still present, its ownership record is missing, or the socket has an active listener, recovery refuses and requires inspection. The desktop app starts its models through this host on all three systems. Multi-GPU workloads have not been qualified on this path.
Owner-editable operation recipes save typed actions next to profiles. Initialize three examples, preview the exact action and its effects, then run it explicitly:
@@ -234,7 +234,7 @@ When `--max-queue` is set, at most **effective slots + max queue** HTTP inferenc
`status --json` and `monitor --once --json` return one versioned JSON object. Remote monitor URLs require verified HTTPS; a local SSH tunnel can use `http://127.0.0.1`. For a read-only monitor key, set `STRATA_TELEMETRY_KEY` on the server and pass the same secret to the client through `--credential-env STRATA_TELEMETRY_KEY` (or another client environment variable). Use a distinct inference API key: the server rejects duplicate keys. The telemetry key authorizes `/v1/telemetry` only; it cannot call inference or management endpoints. The existing `STRATA_API_KEY` also works and has its broader legacy permissions. Keys are never put in URLs or command arguments. The client refuses redirects and responses over 2 MiB. Excessively nested JSON is reported as a malformed telemetry response with the same one-record error contract, rather than a Python traceback.
-The interactive monitor has seven views: Overview, Serving, Requests, Engine, Hardware, Logs and **Connect** (the address, the key and the setup for your coding apps; see [TERMINAL_LAUNCH.md](TERMINAL_LAUNCH.md#after-the-start-the-monitor)). Use keys `1`–`7` or Left/Right to switch views, Tab in Serving to cycle Devices / Requests / Groups at wide and compact widths, `/` to filter requests, Up/Down and Page Up/Down to move the selected row, Home/End to jump to the first/last row, Enter to show or hide read-only request, device or slot detail, Space to freeze the display, `r` to refresh, `t` to cycle theme, `?` or F1 for help, and `q` to detach. Filter input is limited to 80 characters and cleaned of terminal controls and bidi formatters as text arrives; `q` remains ordinary text while the filter has focus. Recent detail contains bounded token and timing summaries, never prompt bodies. A selected request, device or slot stays attached to its reported ID across polls, including when a request moves from live to recent history; detail closes and the TUI says “No longer retained” if that ID disappears. An instance restart clears selection even when a new request reuses the same ID. The 60- and 40-column layouts keep the selected request visible. It never starts or stops the server. `NO_COLOR` or `--color never` uses terminal defaults. `--glyphs ascii` avoids special glyphs. The selected terminal font is controlled by the terminal, not Strata.
+The interactive monitor has seven views: Overview, Serving, Requests, Engine, Hardware, Logs and Connect (the address, the key and the setup for your coding apps; see [TERMINAL_LAUNCH.md](TERMINAL_LAUNCH.md#after-the-start-the-monitor)). Use keys `1`–`7` or Left/Right to switch views, Tab in Serving to cycle Devices / Requests / Groups at wide and compact widths, `/` to filter requests, Up/Down and Page Up/Down to move the selected row, Home/End to jump to the first/last row, Enter to show or hide read-only request, device or slot detail, Space to freeze the display, `r` to refresh, `t` to cycle theme, `?` or F1 for help, and `q` to detach. Filter input is limited to 80 characters and cleaned of terminal controls and bidi formatters as text arrives; `q` remains ordinary text while the filter has focus. Recent detail contains bounded token and timing summaries, never prompt bodies. A selected request, device or slot stays attached to its reported ID across polls, including when a request moves from live to recent history; detail closes and the TUI says “No longer retained” if that ID disappears. An instance restart clears selection even when a new request reuses the same ID. The 60- and 40-column layouts keep the selected request visible. It never starts or stops the server. `NO_COLOR` or `--color never` uses terminal defaults. `--glyphs ascii` avoids special glyphs. The selected terminal font is controlled by the terminal, not Strata.
Remote model, device and request labels are stripped of terminal controls and bidi formatters before display. The stdlib renderer clips by terminal display cells and pads fixed columns by the same measure; local cases cover combining marks, CJK names, regional-indicator flags and joined emoji at 110, 80, 60 and 40 columns. Terminal implementations can differ on ambiguous-width characters, so the renderer uses the terminal's narrow default for them.
diff --git a/docs/TROUBLESHOOTING.md b/docs/TROUBLESHOOTING.md
index a9cf9e7a1..80bf648e7 100644
--- a/docs/TROUBLESHOOTING.md
+++ b/docs/TROUBLESHOOTING.md
@@ -6,7 +6,7 @@ error messages, with the older engine fixes, is in the [details](DETAILS.md#trou
## The desktop app and the Terminal
**A warning says Strata "couldn't find an engine for this computer".**
-The engine is the part that talks to your graphics card. This is a warning, not a refusal: Strata still downloads the
+The engine is the part that talks to your graphics card. Strata only warns and still downloads the
model, and the model files stay on your computer, but the model may not run until the right engine is installed. The usual
reasons are that no engine is installed yet and the update source has none for your card, that the engine found is for
another make of card, or that the graphics driver is too old. In the app, open Settings, then Engine. In the Terminal, the
diff --git a/docs/release-notes-0.2.0.md b/docs/release-notes-0.2.0.md
index 28c9ffedf..7ed27c929 100644
--- a/docs/release-notes-0.2.0.md
+++ b/docs/release-notes-0.2.0.md
@@ -7,7 +7,7 @@
- **The Strata app** for Windows (installer), Linux (AppImage) and Mac (Apple silicon, macOS 15 or newer, `.dmg`). Chat with a coding agent that asks before it acts (Manual) or works freely (Bypass), web search, `/loop`, pictures pasted or dropped into the chat, a Models page, a Storage page, a choice of which graphics cards a model uses, two models on two GPUs, sharing with other devices (needs an API key first) and setup of your coding apps.
- **The Terminal**, one line to install (`curl -fsSL https://github.com/Niko1221/Strata/releases/latest/download/install.sh | sh`, `irm https://github.com/Niko1221/Strata/releases/latest/download/install.ps1 | iex` on Windows). `strata launch` is the guided flow (choose a model, tune it, serve it, watch it) and also runs without screens from flags; `strata connect` writes and checks the setup of OpenCode, Codex CLI, Claude Code, oh-my-pi, pi, Hermes Agent and OpenClaw; `strata models`, `strata update`, `strata doctor` and `strata bench run` are there too. The monitor `strata launch` opens has seven views, `1`-`7`, the seventh is Connect. See `docs/TERMINAL_LAUNCH.md`.
- **The engine**, the part that talks to your card, is a separate package per platform and card. The installers carry one, so a new install starts a model without a download. Later engines come from a feed that is signed (Ed25519) and checked file by file, test-started before the switch and rolled back on any failure; models and saved tunings are never replaced. Update with `strata update` or Settings > Update. See `packaging/UPDATES.md` and `SECURITY.md`.
-- **No engine found is a warning, not a refusal.** `strata launch` and the app say in plain words what is missing and what to do next, and still download the model.
+- **A missing engine is a warning.** `strata launch` and the app say in plain words what is missing and what to do next, and still download the model.
- Starting a model and serving never download anything; downloads are one explicit step.
**Pictures are on by default on NVIDIA and AMD cards.** The picture reader runs on the CPU for a card under 24 GB (a few seconds per picture, nothing taken from the card) and on the graphics card from 24 GB. NVIDIA, AMD and Intel follow the same rule; `--vision gpu` / `--vision cpu` / off still choose per model. AMD now has a GPU reader with an automatic CPU fallback (the server prints one line when the GPU reader cannot start). Measured on a Radeon AI PRO R9700: a 260-token picture 57 ms on the GPU reader against 3.13 s on the CPU, a 972-token picture 0.40 s against 14.1 s; the GPU reader keeps 1,470 MiB of VRAM. On a Radeon 780M 0.40 s against 2.42 s. Gyro and the NVFP4 models read pictures too (Gyro needs engine 0.1.45.1 or newer). The Mac app has no picture reader yet.
@@ -19,12 +19,12 @@
Every speed number below is a median of interleaved pairs against the previous default on the same box (a fresh engine per run); "pairs faster" counts the pairs the new default won. Each default has an off switch.
- **NVIDIA RTX 30 / A-series (sm_86) and Pascal (sm_60): decode.** A per-architecture table of nine (sm_86) and seven (sm_60) switches, picked by removing one at a time and keeping only those that cost speed when removed. RTX 3060 12 GB 60.36 to 66.52 tok/s (+10.2%, 8 of 8 pairs), RTX A4000 16 GB 51.16 to 57.82 (+13.0%, 8 of 8), Tesla P100 16 GB 36.56 to 41.76 (+14.2%, 8 of 8). Same answers: the fixed-flag identity gate (10 prompts, 4K and 20K exact prompt checks) is identical on all three. `STRATA_SM86_PERF=0` / `STRATA_SM60_PERF=0` turn the table off. We left a prompt chunk above 8192 off on NVIDIA (not bit-exact on CUDA) and `STRATA_ROUTE_PRIOR=0.5` opt-in (KL 0.0126 against a run-to-run floor of 0.0012).
-- **AMD RX 9000 / R9700 (gfx12), Linux: prompt reading.** The WMMA prompt attention and the fused Q2_0 prompt experts are on. On an R9700 32 GB, Q2_0: a 4K prompt 1,151 to 2,286 tok/s and a 20K prompt 1,384 to 2,751 (1.99x, 5 of 5 pairs). IQ3_XXS: 4K 1,369 to 1,436 (1.05x), 20K 1,576 to 1,997 (1.27x). Decode is untouched (0.999 like for like, same tokens). **The text changes slightly**: teacher-forced KL 0.0003 (28 prompts) and 0.0015 (long documents), task suites equal; with both on: HumanEval 155/164, reasoning 27/30, agent 6/6, needles 6/6, tool calls 5/5. `STRATA_HIP_WMMA=0` and `STRATA_PF_FUSED=0` turn them off. **On Windows they stay opt-in** until tested there.
+- **AMD RX 9000 / R9700 (gfx12), Linux: prompt reading.** The WMMA prompt attention and the fused Q2_0 prompt experts are on. On an R9700 32 GB, Q2_0: a 4K prompt 1,151 to 2,286 tok/s and a 20K prompt 1,384 to 2,751 (1.99x, 5 of 5 pairs). IQ3_XXS: 4K 1,369 to 1,436 (1.05x), 20K 1,576 to 1,997 (1.27x). Decode is untouched (0.999 like for like, same tokens). The text changes slightly: teacher-forced KL 0.0003 (28 prompts) and 0.0015 (long documents), task suites equal; with both on: HumanEval 155/164, reasoning 27/30, agent 6/6, needles 6/6, tool calls 5/5. `STRATA_HIP_WMMA=0` and `STRATA_PF_FUSED=0` turn them off. On Windows they stay opt-in until tested there.
- **Intel Arc A-series (Alchemist): prompt reading.** The XMX path (`STRATA_PF_XMX=2`) is the default. On an A750, 4K prompt 79.5 to 113.0 tok/s; by length (previous default / new) 1K 35.1 / 46.4, 2K 61.7 / 83.1, 6K 66.3 / 114.8. The logits equal those of the old opt-in; `STRATA_PF_XMX=0` brings back the previous default bit for bit. Not for Battlemage (it loses 7-11% there).
- **Intel Arc A750: a prompt deadlock is fixed.** A prompt of about 5,000 tokens could hang the engine in the driver on pageable memory (the engine stopped, HTTP 503); the reproduction hung 6 of 6 before, completes 3 of 3 now, with identical first-token logits and needles 9 of 9 twice.
- **Intel Arc B-series (Battlemage, B70): int8 experts for prompt reading.** On by default: 1K prompt 648 to 785 tok/s (+21%), 4K 1,264 to 1,419 (+12%), same output; `STRATA_PREFILL_MMQ=0` turns it off. Setup now picks an 8192 prompt chunk for a Battlemage card of 24 GB or more with a context over 32K: 16K prompt 1,222 to about 1,350 tok/s (+10-11%), 32K 1,187 to about 1,370 (+15-16%), decode unchanged.
- **Linux, models read in place from the GGUF: RAM and SSD.** Zero-copy page-cached experts, I/O prefetch with a gate that checks it pays, and a RAM check that counts the container's clean page cache. On a 12 GB card emulated on a 72 GB card with the RAM limited by cgroup (Q2_0, three pairs): with 32 GB decode 1.70x / 2.56x / 2.88x, prompts 1.54x / 1.91x; with 16 GB decode 1.16x / 1.25x / 1.25x, prompts 1.63x / 1.68x; with an 18 GiB RAM tier decode 1.16x / 1.10x / 1.30x. On a real RTX 3060 the zero-copy part alone gave decode 1.14x / 1.44x / 1.47x. The gates are identical (Q2_0 and IQ3_S, with and without `--mmap-experts`). Packs in `experts.bin` keep the old behaviour. Off: `STRATA_COLD_OVERLAP=0`, `STRATA_IO_PREFETCH=0`, `STRATA_IO_PF_GATE=0`.
-- **The prompt chunk now accounts for RAM and SSD (file tier, Linux).** Where today's chunk would make the file set bigger than the page cache, the engine picks a smaller one. Measured on an RTX 3060 with the SATA SSD **emulated** (cgroup RAM limit and a 450 MB/s read limit, five pairs): a 20K prompt 334 to 822 tok/s (2.46x), a 29K prompt 419 to 910 (2.17x), drive reads for the 29K prompt 29.5 GB to 0.6 GB, decode after the prompt 28.6 to 45.7; a 4K prompt unchanged. Where the file set already fits the page cache (the 5070 test PC) it changes nothing. Tested on CUDA only. `STRATA_PREFILL_HOST_PLAN=0` turns it off.
+- **The prompt chunk now accounts for RAM and SSD (file tier, Linux).** Where today's chunk would make the file set bigger than the page cache, the engine picks a smaller one. Measured on an RTX 3060 with the SATA SSD emulated (cgroup RAM limit and a 450 MB/s read limit, five pairs): a 20K prompt 334 to 822 tok/s (2.46x), a 29K prompt 419 to 910 (2.17x), drive reads for the 29K prompt 29.5 GB to 0.6 GB, decode after the prompt 28.6 to 45.7; a 4K prompt unchanged. Where the file set already fits the page cache (the 5070 test PC) it changes nothing. Tested on CUDA only. `STRATA_PREFILL_HOST_PLAN=0` turns it off.
- **CPU worker counts follow the physical cores** (host socket, container quota, cap 32), and setup no longer writes half the logical threads (it wrote 128 on a 256-thread machine). Too many workers cost real speed: with as many workers as logical threads decode fell to 0.65-0.73 of the default on a 6-core and a 32-core box. Speed at the new defaults equals the old within noise (R9700, 31 to 30 workers: prompt 1.005-1.007, decode 1.000-1.007). A configured count above the plan still runs, with a warning.
## Fixes from the owner's tester reports
@@ -53,7 +53,7 @@ Every speed number below is a median of interleaved pairs against the previous d
- **R9700:** about half of the fresh engines with MTP read prompts 10-23% slower than the others (4K: about 1,350 instead of 1,700 tok/s). Not seen on NVIDIA; the cause is not found.
- **Docker images are not published yet;** build them yourself (`docs/DOCKER.md`).
- **`strata code` is not available** in the Terminal; use `strata connect` for coding apps (the app has its own agent).
-- The standard benchmark protocol is a **draft** (`docs/BENCHMARK_PROTOCOL.md`); `strata bench run` and `BENCH.bat` do not implement it.
+- The standard benchmark protocol is a draft (`docs/BENCHMARK_PROTOCOL.md`); `strata bench run` and `BENCH.bat` do not implement it.
Thanks to Laurent Zuijdwijk (the Gyro kernels for CPU and the trellis types), ATIVX928 and NoxestNoz (Volta work) and everyone who tested the builds and sent reports.