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++

launch: text previews of every screen, advanced table widths

Niko12214e6a1b46bead
diff --git a/docs/terminal-previews/launch-monitor-connect-hermes.txt b/docs/terminal-previews/launch-monitor-connect-hermes.txt
new file mode 100644
index 000000000..2806209f4
--- /dev/null
+++ b/docs/terminal-previews/launch-monitor-connect-hermes.txt
@@ -0,0 +1,39 @@
+# strata launch preview: the Connect view, Hermes selected (its limits in plain words)
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Connect http://127.0.0.1:8080 [LIVE]
+ 1 Overview 2 Serving 3 Requests 4 Engine 5 Hardware 6 Logs [7 Connect]
+ ────────────────────────────────────────────────────────────────────────────────────────────────────
+ Connect an app URL http://127.0.0.1:8080/v1 model qwen3.8-flash-next-iq3_xxs key sk-strata-•••• (k show)
+ also reachable at http://192.168.88.21:8080/v1
+
+ App Works with Status
+ OpenCode chat + tools tested ✓
+ Codex CLI responses + tools tested ✓
+ Claude Code messages + tools tested ✓
+ oh-my-pi (omp) chat + tools tested ✓
+ pi (pi.dev) chat + tools tested ✓
+ › Hermes Agent chat + tools tested ✓
+ OpenClaw chat + tools tested ✓
+ DeepSeek Harness the Strata app tested ✓
+
+ Hermes Agent setup ~/.hermes/config.yaml
+ model:
+ default: qwen3.8-flash-next-iq3_xxs
+ provider: lmstudio
+ base_url: http://127.0.0.1:8080/v1
+ context_length: 32768
+ compression:
+ enabled: true
+ ... 5 more lines (c copies all)
+ Good to know
+ - WARNING: Hermes is built for 64K or more. At 32K it works only through the workaround above.
+ - Hermes refuses a window under 64,000 tokens unless the provider is lmstudio with an explicit context_length: the
+ config uses that, and auxiliary.compression.context_length 65536 for its summariser (the same floor).
+ - Its system prompt and 20 tools take about 12.9K tokens (40% of a 32K window): about 19K is left for the
+ conversation.
+ - Its compaction threshold is floored at its own 64K minimum, so it compacts later than 15K (around 20K); it still
+
+
+
+ c copy w write it for me (keeps a backup) h check it 1-7 views q quit (asks: keep serving / stop)
diff --git a/docs/terminal-previews/launch-monitor-connect-opencode.txt b/docs/terminal-previews/launch-monitor-connect-opencode.txt
new file mode 100644
index 000000000..40ea89379
--- /dev/null
+++ b/docs/terminal-previews/launch-monitor-connect-opencode.txt
@@ -0,0 +1,39 @@
+# strata launch preview: the monitor's Connect view (key 7), OpenCode selected
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Connect http://127.0.0.1:8080 [LIVE]
+ 1 Overview 2 Serving 3 Requests 4 Engine 5 Hardware 6 Logs [7 Connect]
+ ────────────────────────────────────────────────────────────────────────────────────────────────────
+ Connect an app URL http://127.0.0.1:8080/v1 model qwen3.8-flash-next-iq3_xxs key sk-strata-•••• (k show)
+ also reachable at http://192.168.88.21:8080/v1
+
+ App Works with Status
+ › OpenCode chat + tools tested ✓
+ Codex CLI responses + tools tested ✓
+ Claude Code messages + tools tested ✓
+ oh-my-pi (omp) chat + tools tested ✓
+ pi (pi.dev) chat + tools tested ✓
+ Hermes Agent chat + tools tested ✓
+ OpenClaw chat + tools tested ✓
+ DeepSeek Harness the Strata app tested ✓
+
+ OpenCode setup ~/.config/opencode/opencode.json
+ {
+ "$schema": "https://opencode.ai/config.json",
+ "model": "strata/qwen3.8-flash-next-iq3_xxs",
+ "small_model": "strata/qwen3.8-flash-next-iq3_xxs",
+ "autoupdate": false,
+ "compaction": {
+ "auto": true,
+ "buffer": 13672
+ ... 25 more lines (c copies all)
+ Good to know
+ - The setup needs limit.input (strata connect sets it to 28672); without it OpenCode never compacts its
+ conversation.
+ - Its first start downloads a plug-in from npm (internet needed once; the first run can take over a minute).
+ - In a script or over ssh, run it as: opencode run "..." < /dev/null (it waits for typing otherwise).
+ - Its system prompt and tools take about 7.4K tokens of the window.
+
+
+
+ c copy w write it for me (keeps a backup) h check it 1-7 views q quit (asks: keep serving / stop)
diff --git a/docs/terminal-previews/launch-monitor-overview.txt b/docs/terminal-previews/launch-monitor-overview.txt
new file mode 100644
index 000000000..1eb105e03
--- /dev/null
+++ b/docs/terminal-previews/launch-monitor-overview.txt
@@ -0,0 +1,39 @@
+# strata launch preview: the monitor's new Overview (view 1)
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Serving http://127.0.0.1:8095 IQ3_S 128K ● Reading prompt · 97% [LIVE]
+ [1 Overview] 2 Serving 3 Requests 4 Engine 5 Hardware 6 Logs 7 Connect
+ ──────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Model state Idle [Reading] Generating Queued Error 35,610 / 36,742 tokens · 97%
+ █████████████████████████████████████████████████████████████████████████████████████████████████████████░░░
+
+ ╭─ Speed ────────────────╮ ╭─ GPU load ─────────────╮ ╭─ VRAM ─────────────────╮ ╭─ GPU temp ─────────────╮
+ │ 820.0 tok/s │ │ 88 % │ │ 11.2 / 12 GB │ │ 54 °C │
+ │ reading the prompt │ │ RTX 5070 │ │ 1,644 experts cached │ │ │
+ │ ▂▆▁▄█▂▆▁▄█▂▆▁▄█▂▆▁▄█▂▆ │ │ ▇▇▇▇█▇▇▇▇█▇▇▇▇█▇▇▇▇█▇▇ │ │ ██████████████████████ │ │ ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ │
+ ╰────────────────────────╯ ╰────────────────────────╯ ╰────────────────────────╯ ╰────────────────────────╯
+ ╭─ Power ────────────────╮ ╭─ PCIe ─────────────────╮ ╭─ CPU ──────────────────╮ ╭─ Disk read ────────────╮
+ │ 114 W │ │ Gen4 x16 │ │ 14 % │ │ 0.1 MB/s │
+ │ of 250 W limit │ │ to GPU 18,237 MB/s │ │ 6 cores · 12 threads │ │ write 0.0 MB/s │
+ │ ██████████████████████ │ │ ▆█▅▇█▆█▅▇█▆█▅▇█▆█▅▇█▆█ │ │ ▁▂▁▂▂▁▂▁▂▂▁▂▁▂▂▁▂▁▂▂▁▂ │ │ ▂▆▁▄█▂▆▁▄█▂▆▁▄█▂▆▁▄█▂▆ │
+ ╰────────────────────────╯ ╰────────────────────────╯ ╰────────────────────────╯ ╰────────────────────────╯
+ Context fill 36K / 128K 28% Recent requests
+ ████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ TIME STATUS PROMPT REUSED OUTPUT TOK/S DURATION
+ Experts in VRAM 1,644 · 3.2 GB 02:14:56 Done 17,183 0 19,511 50.5 429.8 s
+ ██████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 02:14:52 Done 626 0 11 26.8 4.0 s
+ System RAM 58.5 / 63 GB
+ █████████████████████████████████████████░░░
+
+ 1-7 views / filter Space freeze r refresh t theme b open in browser ? help q quit
+
+
+
+
+
+
+
+
+
+
+
+ Read only | q detaches without stopping Strata
diff --git a/docs/terminal-previews/launch-monitor-quit.txt b/docs/terminal-previews/launch-monitor-quit.txt
new file mode 100644
index 000000000..9bdb4b043
--- /dev/null
+++ b/docs/terminal-previews/launch-monitor-quit.txt
@@ -0,0 +1,27 @@
+# strata launch preview: q asks: keep serving or stop
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ┌────────────────────────────────────────────────────────────────────┐
+ │ │
+ │ Leave the monitor? │
+ │ Keeping it serving lets your apps go on using the model. │
+ │ Stopping unloads the model and frees the graphics memory. │
+ │ │
+ │ › Keep serving in the background │
+ │ Stop serving and free the memory │
+ │ Stay in the monitor │
+ │ Up/Down choose Enter confirm Esc cancel │
+ │ │
+ └────────────────────────────────────────────────────────────────────┘
diff --git a/docs/terminal-previews/launch-overview-ascii.txt b/docs/terminal-previews/launch-overview-ascii.txt
new file mode 100644
index 000000000..effe95495
--- /dev/null
+++ b/docs/terminal-previews/launch-overview-ascii.txt
@@ -0,0 +1,26 @@
+# strata monitor Overview in the classic Windows console mode (--glyphs ascii, or detected): no characters Consolas lacks
+
+STRATA Serving http://127.0.0.1:8080 IQ3_S 128K * Generating a reply · 3% [LIVE]
+[1 Overview] 2 Serving 3 Requests 4 Engine 5 Hardware 6 Logs 7 Connect
+--------------------------------------------------------------------------------------------------------------
+Model state Idle Reading [Generating] Queued Error 120 / 4,096 tokens · 3%
+###.........................................................................................................
+
+┌─ Speed ────────────────┐ ┌─ GPU load ─────────────┐ ┌─ VRAM ─────────────────┐ ┌─ GPU temp ─────────────┐
+│ 50.4 tok/s │ │ 88 % │ │ 11.2 / 12 GB │ │ 54 C │
+│ now │ │ RTX 5070 │ │ 1,644 experts cached │ │ │
+│ :*.=@:*.=@:*.=@:*.=@:* │ │ ####@####@####@####@## │ │ @@@@@@@@@@@@@@@@@@@@@@ │ │ ====================== │
+└────────────────────────┘ └────────────────────────┘ └────────────────────────┘ └────────────────────────┘
+┌─ Power ────────────────┐ ┌─ PCIe ─────────────────┐ ┌─ CPU ──────────────────┐ ┌─ Disk read ────────────┐
+│ 114 W │ │ Gen4 x16 │ │ 14 % │ │ 0.1 MB/s │
+│ of 250 W limit │ │ to GPU 18,237 MB/s │ │ 6 cores · 12 threads │ │ write 0.0 MB/s │
+│ @@@@@@@@@@@@@@@@@@@@@@ │ │ *@+#@*@+#@*@+#@*@+#@*@ │ │ .:.::.:.::.:.::.:.::.: │ │ :*.=@:*.=@:*.=@:*.=@:* │
+└────────────────────────┘ └────────────────────────┘ └────────────────────────┘ └────────────────────────┘
+Context fill 36K / 128K 28% Recent requests
+############................................ TIME STATUS PROMPT REUSED OUTPUT TOK/S DURATION
+Experts in VRAM 1,644 · 3.2 GB 02:14:58 Done 17,183 0 19,511 50.5 429.8 s
+##############.............................. 02:14:54 Done 626 0 11 26.8 4.0 s
+System RAM 58.5 / 63 GB
+#########################################...
+
+1-7 views / filter Space freeze r refresh t theme b open in browser ? help q quit
diff --git a/docs/terminal-previews/launch-run-progress.txt b/docs/terminal-previews/launch-run-progress.txt
new file mode 100644
index 000000000..0b895dc97
--- /dev/null
+++ b/docs/terminal-previews/launch-run-progress.txt
@@ -0,0 +1,36 @@
+# strata launch preview: the progress screen while it downloads and tunes
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Launch Demo
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ ✓ Engine Engine 0.1.45 (cuda) is installed.
+ ✓ Model Model is ready.
+ ✓ Settings Settings are saved.
+ > Tuning █████████████████████████░░░░░░░░░░░░░░░ 62% Tuning
+ Starting
+ Access
+
+
+ Tuning measurement 5 of 8 about 7 min left
+ Default settings 83.0 tokens/s
+ Keep more context on GPU 78.5 tokens/s
+ Longer prompt chunks 85.1 tokens/s
+ Close other heavy programs while it measures. You can stop it any time: nothing changes until it is done.
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Strata is getting everything ready. This can take a while the first time (the model is large). Stopping is
+ safe: files already downloaded are kept and a download continues where it stopped.
+ s stop q quit
diff --git a/docs/terminal-previews/launch-start.txt b/docs/terminal-previews/launch-start.txt
new file mode 100644
index 000000000..2b48a3af6
--- /dev/null
+++ b/docs/terminal-previews/launch-start.txt
@@ -0,0 +1,36 @@
+# strata launch preview: start screen with saved presets
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Launch RTX 5070 12 GB · 64 GB RAM · 93 GB free
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Start from a saved preset
+
+ › IQ3_XXS 128K - Full speed IQ3_XXS 128K 1 request tuned 10-16 last used 10-16
+ Network serving Q2_0 32K 1 request tuned 10-15 last used 10-15
+ Coder 64K - agents IQ1_M 64K 2 requests not tuned last used 10-14
+ Set up from the beginning
+ Use model files I already have
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ This computer: RTX 5070 12 GB · 64 GB RAM · 93 GB free. Preset: model, context, parallel requests, tuned
+ settings, access. Shared with the Strata app.
+ ↑↓ choose Enter launch e edit preset d delete ? help q quit
diff --git a/docs/terminal-previews/launch-step1-model.txt b/docs/terminal-previews/launch-step1-model.txt
new file mode 100644
index 000000000..ada5cc2ad
--- /dev/null
+++ b/docs/terminal-previews/launch-step1-model.txt
@@ -0,0 +1,36 @@
+# strata launch preview: step1-model
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Set up Step 1 of 9: Model
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Which model? This computer: RTX 5070 12 GB · 64 GB RAM · 93 GB free
+ Model Download Needs (graphics + RAM) On this computer
+ Qwen3.8-Flash-Next Q2_0 66 GB 12 GB + 27 GB fits
+ Qwen3.8-Flash-Next IQ2_XS 68 GB 12 GB + 28 GB fits
+ › Qwen3.8-Flash-Next IQ3_XXS 76 GB 12 GB + 36 GB fits (recommended)
+ Qwen3.8-Flash-Next IQ3_S 84 GB 12 GB + 43 GB fits
+ Swift 1.5 IQ2_XS 68 GB 12 GB + 28 GB fits
+ Swift 1.5 IQ3_XXS 76 GB 12 GB + 36 GB fits
+ Swift 1.5 IQ3_S 84 GB 12 GB + 43 GB fits
+ Qwen3.8-Flash-Next Coder IQ1_M 58 GB 12 GB + 16 GB fits
+ Qwen3.8-Flash-Next (Unsloth) UD-IQ4_XS 94 GB 12 GB + 52 GB fits, part read from SSD
+ Qwen3.8-Flash-Next (Unsloth) UD-Q4_K_XL 111 GB 12 GB + 54 GB fits, part read from SSD
+ Qwen3.8-Flash-Next (Unsloth) UD-Q5_K_XL 158 GB 12 GB + 54 GB tight: much read from SSD
+ Qwen3.8-Flash-Next (Unsloth) UD-Q6_K_XL 169 GB 12 GB + 54 GB tight: much read from SSD
+ Qwen3.8-Flash-Next (Unsloth) Q8_0 188 GB 12 GB + 54 GB tight: much read from SSD
+ Qwen3.8-Flash-Next (Gyro) Gyro-S 58 GB 12 GB + 20 GB fits
+ Qwen3.8-Flash-Next (Gyro) Gyro-M 92 GB 12 GB + 27 GB fits
+ Qwen3.8-Flash-Next (NVFP4) NVFP4 128 GB 12 GB + 54 GB does not fit (can still try)
+ Qwen3.8-Flash-Next (NVFP4) NVFP4-NVIDIA 128 GB 12 GB + 54 GB does not fit (can still try)
+ Qwen3.8-Flash-Next (NVFP4) NVFP4-QAD 102 GB 12 GB + 54 GB does not fit (can still try)
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Qwen3.8-Flash-Next IQ3_XXS: 3-bit i-quant, better quality, slower (more CPU work per token).
+ The n-gram table stays on your SSD and is not counted above.
+ ↑↓ choose Enter next o use my own files Esc back ? help
diff --git a/docs/terminal-previews/launch-step2-folder.txt b/docs/terminal-previews/launch-step2-folder.txt
new file mode 100644
index 000000000..34183488a
--- /dev/null
+++ b/docs/terminal-previews/launch-step2-folder.txt
@@ -0,0 +1,36 @@
+# strata launch preview: step2-folder
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Set up Step 2 of 9: Folder
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Where should the model files be stored?
+ Needs about 84 GB (the download plus working room).
+ Free space in that folder's drive: 93 GB
+ ┌──────────────────────────────────────────────────────────────────────────────────────────────────────────┐
+ │ C:\Users\AI-Server\Documents\Strata │
+ └──────────────────────────────────────────────────────────────────────────────────────────────────────────┘
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Models are large files. Strata keeps them in one folder; the default is Strata's own data folder. Press
+ Enter to keep it, or type another folder (it is created if it does not exist).
+ Type to change the folder Enter next Esc back ? help
diff --git a/docs/terminal-previews/launch-step3-context.txt b/docs/terminal-previews/launch-step3-context.txt
new file mode 100644
index 000000000..3ce4d5d20
--- /dev/null
+++ b/docs/terminal-previews/launch-step3-context.txt
@@ -0,0 +1,36 @@
+# strata launch preview: step3-context
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Set up Step 3 of 9: Context
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ How much text should it remember at once?
+ 8K about 6,100 words +0.1 GB
+ 32K about 25,000 words +0.4 GB recommended
+ 64K about 49,000 words +0.9 GB
+ › 128K about 98,000 words +1.8 GB
+ 200K about 150,000 words +2.8 GB tight on this computer
+ 256K about 200,000 words +3.6 GB tight on this computer
+ 384K about 290,000 words +5.4 GB experimental: rope scaling
+ 512K about 390,000 words +7.2 GB experimental: rope scaling
+ Custom…
+ Conversation memory: 8-bit (recommended: what the published numbers were measured with) press m to change
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ More context lets agents read bigger projects. It uses more memory and the first reply of a long chat is
+ slower. The memory shown is only the extra for the conversation itself.
+ ↑↓ choose Enter next m memory format Esc back ? help
diff --git a/docs/terminal-previews/launch-step4-pictures.txt b/docs/terminal-previews/launch-step4-pictures.txt
new file mode 100644
index 000000000..02dd47cbe
--- /dev/null
+++ b/docs/terminal-previews/launch-step4-pictures.txt
@@ -0,0 +1,36 @@
+# strata launch preview: step4-pictures
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Set up Step 4 of 9: Pictures
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Should the model read pictures?
+ › No pictures
+ Yes, read pictures
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Pictures add a 0.9 GB download. On a card under 24 GB the picture reader runs on the CPU (a few seconds per
+ picture, nothing taken from the graphics card). On a bigger card it runs on the GPU and keeps about 1.4 GB
+ free, costing a few % of text speed. Say No if you only chat in text.
+ ↑↓ choose Enter next Esc back ? help q quit
diff --git a/docs/terminal-previews/launch-step5-parallel.txt b/docs/terminal-previews/launch-step5-parallel.txt
new file mode 100644
index 000000000..a417ab008
--- /dev/null
+++ b/docs/terminal-previews/launch-step5-parallel.txt
@@ -0,0 +1,36 @@
+# strata launch preview: step5-parallel
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Set up Step 5 of 9: Parallel
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ How many requests at the same time?
+ › Fit to my graphics card (1) recommended
+ 1 request
+ 2 requests may not fit: slower answers
+ 3 requests may not fit: slower answers
+ 4 requests may not fit: slower answers
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Your card has 12 GB of graphics memory, which fits 1 at once by Strata's rule (12 GB: 1, 20 GB: 2, 24 GB: 3,
+ 32 GB or more: 4). One person: pick 1. A team or several coding agents: more. More than fits still starts,
+ only slower.
+ ↑↓ choose Enter next Esc back ? help q quit
diff --git a/docs/terminal-previews/launch-step6-tuning.txt b/docs/terminal-previews/launch-step6-tuning.txt
new file mode 100644
index 000000000..f84c84be7
--- /dev/null
+++ b/docs/terminal-previews/launch-step6-tuning.txt
@@ -0,0 +1,36 @@
+# strata launch preview: step6-tuning
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Set up Step 6 of 9: Tuning
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Tune for the best speed on this computer?
+ › Quick about 20 minutes checks the most likely gains
+ Thorough up to 2 hours tries more settings, measures each one several times
+ Skip use the default settings (already good on most computers)
+
+ [ ] Allow settings that change answers very slightly (same quality) press a to change
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Tuning is optional. It tries single changes against your settings, keeps only gains above the noise, and
+ saves the result as a preset the Strata app can use too. Close other heavy programs while it measures. If
+ your computer is already at its best it says so and changes nothing.
+ ↑↓ choose Enter next a answer-changing settings on/off Esc back ? help
diff --git a/docs/terminal-previews/launch-step7-advanced.txt b/docs/terminal-previews/launch-step7-advanced.txt
new file mode 100644
index 000000000..29c64c474
--- /dev/null
+++ b/docs/terminal-previews/launch-step7-advanced.txt
@@ -0,0 +1,36 @@
+# strata launch preview: step7-advanced
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Set up Step 7 of 9: Advanced
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ All settings (optional · for tinkerers) / filter x show experimental only
+ Setting Default Tuned Yours What it does
+ › --expert-cache auto auto - experts kept in graphics memory
+ --kv int8 int8 - how the conversation memory is stored
+ --max-context 131072 131072 - memory for the conversation
+ --prefill auto auto - how many prompt tokens are read at on
+ --spec 4 4 - draft tokens per step
+ --native-rope - - - [experimental] experimental pinned ro
+ STRATA_PIN_BY_NEED off off - Pin memory only when needed
+ + Add a setting by name… for flags of newer engine versions
+ Command that will run
+ component:strata-engine --max-context 131072 --kv int8 --prefill auto --expert-cache auto --spec 4
+ Same as: strata launch --model iq3_xxs --context 128k --vision off --tune quick --access local --port 8080
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ --expert-cache: experts kept in graphics memory
+ ↑↓ choose Enter edit Del reset x experimental / filter a add by name c copy command n next
diff --git a/docs/terminal-previews/launch-step8-access.txt b/docs/terminal-previews/launch-step8-access.txt
new file mode 100644
index 000000000..5d324baa5
--- /dev/null
+++ b/docs/terminal-previews/launch-step8-access.txt
@@ -0,0 +1,37 @@
+# strata launch preview: step8-access
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Set up Step 8 of 9: Access
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Who can use it?
+ This computer only http://127.0.0.1:8080 for apps on this computer
+ › My local network http://192.168.88.21:8080 other computers at home or the office
+ Cloudflare tunnel https://<random>.trycloudflare.com from anywhere · cloudflared found
+ Tailscale http://<tailnet address>:8080 Tailscale is not installed
+
+ Port [ 8080 ] press p to change
+ API key [ on ] sk-strata-•••••••• (recommended for anything beyond this computer)
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Other computers on your network can use it at the address shown. Keep the API key on.
+ ↑↓ choose Enter next p port k new key t key on/off Esc back ? help
+ An API key was made for you, because this reaches beyond your computer.
diff --git a/docs/terminal-previews/launch-step9-save.txt b/docs/terminal-previews/launch-step9-save.txt
new file mode 100644
index 000000000..1ea3a054f
--- /dev/null
+++ b/docs/terminal-previews/launch-step9-save.txt
@@ -0,0 +1,36 @@
+# strata launch preview: step9-save
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Set up Step 9 of 9: Save
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Save as preset, then launch
+ Name (type below)
+ Model Qwen3.8-Flash-Next IQ3_XXS
+ Context 128K, memory int8
+ Pictures off
+ Parallel fit to your card (1)
+ Tuning quick (about 20 min)
+ Access My local network, port 8080, API key on
+ Storage Strata's data folder
+
+ ┌──────────────────────────────────────────────────────────────────────────────────────────────────────────┐
+ │ IQ3_XXS 128K │
+ └──────────────────────────────────────────────────────────────────────────────────────────────────────────┘
+ › Save as a preset and launch
+ Launch without saving
+ Save the preset and stop here
+
+
+
+
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Press Enter on the first choice to save and start. The first start may download the model and tune; the
+ monitor opens when it is serving.
+ ↑↓ choose Enter do it Tab rename Esc back ? help
diff --git a/docs/terminal-previews/launch-twocards-start.txt b/docs/terminal-previews/launch-twocards-start.txt
new file mode 100644
index 000000000..b984b5052
--- /dev/null
+++ b/docs/terminal-previews/launch-twocards-start.txt
@@ -0,0 +1,36 @@
+# strata launch preview: start screen with saved presets
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Launch NVIDIA RTX A4000 16 GB x2 · 64 GB RAM · 93 GB free
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Start from a saved preset
+
+ › IQ3_XXS 128K - Full speed IQ3_XXS 128K 1 request tuned 10-16 last used 10-16
+ Network serving Q2_0 32K 1 request tuned 10-15 last used 10-15
+ Coder 64K - agents IQ1_M 64K 2 requests not tuned last used 10-14
+ Set up from the beginning
+ Use model files I already have
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ This computer: NVIDIA RTX A4000 16 GB x2 · 64 GB RAM · 93 GB free. Preset: model, context, parallel
+ requests, tuned settings, access. Shared with the Strata app.
+ ↑↓ choose Enter launch e edit preset d delete ? help q quit
diff --git a/docs/terminal-previews/launch-twocards-step2-cards.txt b/docs/terminal-previews/launch-twocards-step2-cards.txt
new file mode 100644
index 000000000..7ac2beedd
--- /dev/null
+++ b/docs/terminal-previews/launch-twocards-step2-cards.txt
@@ -0,0 +1,36 @@
+# strata launch preview: step2-cards
+# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)
+
+ STRATA Set up Step 2 of 10: Cards
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ Which graphics cards should it use?
+ › Let Strata choose (recommended): 2 cards
+ Only card 0: NVIDIA RTX A4000, 16 GB
+ Only card 1: NVIDIA RTX A4000, 16 GB
+ All of them together, as one
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ ────────────────────────────────────────────────────────────────────────────────────────────────────────────
+ This computer has several graphics cards. One card is simplest; several cards share the model so bigger
+ models fit. Strata's own choice is a good default.
+ ↑↓ choose Enter next Esc back ? help q quit
diff --git a/strata_cli/launch/screens.py b/strata_cli/launch/screens.py
index 796d3fda0..eed754678 100644
--- a/strata_cli/launch/screens.py
+++ b/strata_cli/launch/screens.py
@@ -869,7 +869,7 @@ class AdvancedScreen(Base):
def body(self) -> ComposeResult:
yield Static("", id="advhead", markup=False)
- yield Menu([("Setting", 28), ("Default", 13), ("Tuned", 15), ("Yours", 13), ("What it does", 0)], id="menu", window=8)
+ yield Menu([("Setting", 24), ("Default", 11), ("Tuned", 13), ("Yours", 10), ("What it does", 0)], id="menu", window=8)
yield Static("", id="cmdline", markup=False)
yield Input(id="edit", classes="field", disabled=True)
diff --git a/tests/terminal/launch_previews.py b/tests/terminal/launch_previews.py
new file mode 100644
index 000000000..e94f830a6
--- /dev/null
+++ b/tests/terminal/launch_previews.py
@@ -0,0 +1,153 @@
+"""Writes the text previews of every `strata launch` screen to docs/terminal-previews/launch-*.txt (fake computer: RTX 5070 12 GB, 64 GB RAM).
+
+ python tests/terminal/launch_previews.py
+
+They are what the screens draw at 110 x 34 (the monitor views at 120 x 36), as text, so the design can be read without running anything.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import io
+import os
+import sys
+import time
+from pathlib import Path
+from unittest import mock
+
+ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(ROOT))
+
+from rich.console import Console # noqa: E402
+
+from strata_cli.launch import access, settings as S, store # noqa: E402
+from strata_cli.launch.app import Ctx, LaunchApp # noqa: E402
+from strata_cli.launch.widgets import Glyphs # noqa: E402
+from tests.terminal import test_overview as ov # noqa: E402
+from tests.terminal.launch_support import HW_12, HW_2X16, Home, catalog, text_of # noqa: E402
+
+OUT = ROOT / "docs" / "terminal-previews"
+SIZE = (110, 34)
+
+
+def ctx_for(hw):
+ cat = catalog(hw)
+ comp = cat.computer()
+ return Ctx(catalog=cat, computer=comp, rows=cat.rows(comp.raw), settings=S.LaunchSettings(), glyphs=Glyphs(True))
+
+
+def write(name: str, text: str, note: str) -> None:
+ body = f"# strata launch preview: {note}\n# (text of the screen as drawn at the size named in tests/terminal/launch_previews.py)\n\n" + text.rstrip() + "\n"
+ (OUT / f"launch-{name}.txt").write_text(body, encoding="utf-8")
+ print("wrote", f"launch-{name}.txt")
+
+
+async def keys(pilot, *ks):
+ for k in ks:
+ await pilot.press(k)
+ await pilot.pause(0.15)
+
+
+def run() -> None:
+ OUT.mkdir(parents=True, exist_ok=True)
+ with Home():
+ store.save(S.LaunchSettings(name="IQ3_XXS 128K - Full speed", family="qwen", model="IQ3_XXS", context=131072, parallel=1,
+ tuned_preset="x", tuned_at="2026-10-16", last_used="2026-10-16 09:00:00"))
+ store.save(S.LaunchSettings(name="Coder 64K - agents", family="coder", model="IQ1_M", context=65536, parallel=2, last_used="2026-10-14 09:00:00"))
+ store.save(S.LaunchSettings(name="Network serving", family="qwen", model="Q2_0", context=32768, parallel=1, tuned_preset="y",
+ tuned_at="2026-10-15", last_used="2026-10-15 09:00:00"))
+ base = {"exe": "component:strata-engine", "args": ["--max-context", "131072", "--kv", "int8", "--prefill", "auto", "--expert-cache", "auto",
+ "--spec", "4"]}
+ help_text = (" --max-context N memory for the conversation (default 4096)\n --kv fp16|int8 how the conversation memory is stored\n"
+ " --prefill N how many prompt tokens are read at once\n --expert-cache N experts kept in graphics memory\n"
+ " --spec N draft tokens per step (faster replies)\n --native-rope experimental pinned rotary arithmetic\n")
+ patches = [mock.patch.object(Ctx, "base_config", lambda self: base), mock.patch.object(Ctx, "engine_exe", lambda self: "/engine/strata"),
+ mock.patch("strata_cli.launch.params.engine_help_text", return_value=help_text),
+ mock.patch.object(access, "lan_ips", return_value=["192.168.88.21"]),
+ mock.patch.object(access, "find_cloudflared", return_value="/usr/bin/cloudflared"),
+ mock.patch.object(access, "tailscale_status", return_value={"installed": False, "running": False, "ip": None, "name": None})]
+ for p in patches:
+ p.start()
+
+ async def tour(hw, prefix, steps_after_start):
+ ctx = ctx_for(hw)
+ app = LaunchApp(ctx)
+ async with app.run_test(size=SIZE) as pilot:
+ await pilot.pause(0.3)
+ write(f"{prefix}start", text_of(app), "start screen with saved presets")
+ for name, ks in steps_after_start:
+ await keys(pilot, *ks)
+ write(f"{prefix}{name}", text_of(app), name)
+
+ asyncio.run(tour(HW_12, "", [
+ ("step1-model", ["down", "down", "down", "enter"]),
+ ("step2-folder", ["enter"]),
+ ("step3-context", ["enter", "down", "down"]),
+ ("step4-pictures", ["enter"]),
+ ("step5-parallel", ["enter"]),
+ ("step6-tuning", ["enter"]),
+ ("step7-advanced", ["enter"]),
+ ("step8-access", ["n", "down"]),
+ ("step9-save", ["enter"]),
+ ]))
+ asyncio.run(tour(HW_2X16, "twocards-", [("step2-cards", ["down", "down", "down", "enter", "enter"])]))
+
+ class Runner:
+ def __init__(self, s, **kw):
+ self.s, self.emit = s, kw["emit"]
+
+ def run(self, start=True):
+ from strata_cli.launch.runner import Event, Result
+ self.emit(Event("engine", "Engine 0.1.45 (cuda) is installed.", 100))
+ self.emit(Event("model", "Downloading 41.0 of 75.8 GB", 54.0))
+ time.sleep(1.2)
+ self.emit(Event("model", "Model is ready.", 100))
+ self.emit(Event("config", "Settings are saved.", 100))
+ self.emit(Event("tune", "Tuning", 62.0, data={"round": (5, 8), "minutes_left": 7,
+ "speeds": [{"label": "Default settings", "decode_tok_s": 83.0},
+ {"label": "Keep more context on GPU", "decode_tok_s": 78.5},
+ {"label": "Longer prompt chunks", "decode_tok_s": 85.1}]}))
+ time.sleep(3)
+ return Result("launch-x", ["http://127.0.0.1:8080"], None, "faster", [], "/x")
+
+ async def run_screen():
+ ctx = ctx_for(HW_12)
+ ctx.settings = S.LaunchSettings(name="Demo", family="qwen", model="IQ3_XXS", context=131072, tune="quick")
+ with mock.patch("strata_cli.launch.app.Runner", Runner):
+ app = LaunchApp(ctx)
+ async with app.run_test(size=SIZE) as pilot:
+ await pilot.pause(0.2)
+ app.start_preset("Demo") if False else app.run_launch()
+ await pilot.pause(2.0)
+ write("run-progress", text_of(app), "the progress screen while it downloads and tunes")
+
+ asyncio.run(run_screen())
+
+ # the monitor: Overview, Connect, quit question
+ from strata_cli.launch.monitor_ext import LaunchMonitor
+
+ async def monitor():
+ src = ov.Source(ov.metrics("reading"))
+ info = {"url": "http://127.0.0.1:8080", "urls": ["http://127.0.0.1:8080", "http://192.168.88.21:8080"], "key": "sk-strata-abcdefghijklmn",
+ "model": "qwen3.8-flash-next-iq3_xxs", "context": 32768, "profile": "launch-x", "name": "Demo"}
+ app = LaunchMonitor(src, None, info=info, interval=30)
+ async with app.run_test(size=(120, 36)) as pilot:
+ await pilot.pause(0.6)
+ write("monitor-overview", text_of(app), "the monitor's new Overview (view 1)")
+ await keys(pilot, "7")
+ write("monitor-connect-opencode", text_of(app), "the monitor's Connect view (key 7), OpenCode selected")
+ await keys(pilot, "down", "down", "down", "down", "down")
+ write("monitor-connect-hermes", text_of(app), "the Connect view, Hermes selected (its limits in plain words)")
+ await keys(pilot, "7", "q")
+ write("monitor-quit", text_of(app), "q asks: keep serving or stop")
+
+ asyncio.run(monitor())
+ for p in patches:
+ p.stop()
+ Path(OUT / "launch-overview-ascii.txt").write_text(
+ "# strata monitor Overview in the classic Windows console mode (--glyphs ascii, or detected): no characters Consolas lacks\n\n"
+ + ov.draw(ov.overview.build(ov.metrics("generating")), glyphs="ascii", color=False, connect=True), encoding="utf-8")
+
+
+if __name__ == "__main__":
+ run()