Agent Gen 0.5.37

Unity, live, in one window. Her face moves when she talks, her voice is her own, she hears you, and she becomes whatever you tell her to be.

Download AgentGen-Setup-0.5.37.exe below, run it, pick a drive with room. The first launch asks your age (18+ only), then downloads what she runs on (about 80 GB: Python, PyTorch, her models, her brain).

Needs: Windows 11 64-bit, an NVIDIA GPU with 16 GB VRAM (RTX 30/40/50), 32 GB RAM, 100 GB of free storage for the install folder, speakers and a mic. 18+.

Privacy: nothing you say or show leaves your machine. The one exception is her new web search: the search words and the pages she opens go to DuckDuckGo/Bing and those sites, and every one is listed in the chat. WEB=0 turns it off. See PRIVACY.md.

The installer is not code-signed yet, so Windows SmartScreen may warn: More info → Run anyway.

In this release (0.5.32 – 0.5.37)

• She shows up moving: every fresh start (and Reset) her own video plays as her live body with her lips on top, and one of seven poses from it is her base for outfit changes, never the same one twice running
• She does what you say even when her brain only answers with a mood: "put on a hoodie" or "hands you a hoodie" dresses her, and "is it cold out?" gets the real weather where you are (or wherever you name) before she answers; news, scores and "look it up" get a web search the same way
• Her lips stay on her words: each sentence waits until enough of her face is rendered to play in step with her voice (her face renders below 25 fps, so her lips used to fall behind)
• Age gate at first launch: confirm 18+ and your birth date, or the app closes
• She looks things up: ask about news, scores, weather or a page and she searches the web, reads it and tells you; links show in the chat
• She keeps an eye out: with camera or screen shared she may speak up on her own (you come back, you look upset, you hold something up), never while you talk
• Show camera / Show screen: one still goes with every message so she reacts to you and your screen
• Lighter on the GPU: the image engine unloads when idle, her eyes unload after every look, her brain reserves only the context she uses

Agent Gen 0.5.36

Unity, live, in one window. Her face moves when she talks, her voice is her own, she hears you, and she becomes whatever you tell her to be.

Download AgentGen-Setup-0.5.36.exe below, run it, pick a drive with room. The first launch asks your age (18+ only), then downloads what she runs on (about 80 GB: Python, PyTorch, her models, her brain).

Needs: Windows 11 64-bit, an NVIDIA GPU with 16 GB VRAM (RTX 30/40/50), 32 GB RAM, 100 GB of free storage for the install folder, speakers and a mic. 18+.

Privacy: nothing you say or show leaves your machine. The one exception is her new web search: the search words and the pages she opens go to DuckDuckGo/Bing and those sites, and every one is listed in the chat. WEB=0 turns it off. See PRIVACY.md.

The installer is not code-signed yet, so Windows SmartScreen may warn: More info → Run anyway.

In this release (0.5.32 – 0.5.36)

• She shows up different every time: each fresh start (and Reset) picks one of eight poses from her own video, never the same one twice running, and sometimes she starts already moving with the whole video on her face
• She does what you say even when her brain only answers with a mood: "put on a hoodie" or "hands you a hoodie" dresses her, and "is it cold out?" gets the real weather where you are (or wherever you name) before she answers; news, scores and "look it up" get a web search the same way
• Her lips stay on her words: each sentence waits until enough of her face is rendered to play in step with her voice (her face renders below 25 fps, so her lips used to fall behind)
• Age gate at first launch: confirm 18+ and your birth date, or the app closes
• She looks things up: ask about news, scores, weather or a page and she searches the web, reads it and tells you; links show in the chat
• She keeps an eye out: with camera or screen shared she may speak up on her own (you come back, you look upset, you hold something up), never while you talk
• Show camera / Show screen: one still goes with every message so she reacts to you and your screen
• Lighter on the GPU: the image engine unloads when idle, her eyes unload after every look, her brain reserves only the context she uses

Agent Gen 0.5.35

Unity, live, in one window. Her face moves when she talks, her voice is her own, she hears you, and she becomes whatever you tell her to be.

Download AgentGen-Setup-0.5.35.exe below, run it, pick a drive with room. The first launch asks your age (18+ only), then downloads what she runs on (about 80 GB: Python, PyTorch, her models, her brain).

Needs: Windows 11 64-bit, an NVIDIA GPU with 16 GB VRAM (RTX 30/40/50), 32 GB RAM, 100 GB of free storage for the install folder, speakers and a mic. 18+.

Privacy: nothing you say or show leaves your machine. The one exception is her new web search: the search words and the pages she opens go to DuckDuckGo/Bing and those sites, and every one is listed in the chat. WEB=0 turns it off. See PRIVACY.md.

The installer is not code-signed yet, so Windows SmartScreen may warn: More info → Run anyway.

In this release (0.5.32 – 0.5.35)

• She does what you say even when her brain only answers with a mood: "put on a hoodie" or "hands you a hoodie" dresses her, and "is it cold out?" gets the real weather where you are (or wherever you name) before she answers; news, scores and "look it up" get a web search the same way
• Her lips stay on her words: each sentence waits until enough of her face is rendered to play in step with her voice (her face renders below 25 fps, so her lips used to fall behind)
• Age gate at first launch: confirm 18+ and your birth date, or the app closes
• She looks things up: ask about news, scores, weather or a page and she searches the web, reads it and tells you; links show in the chat
• She keeps an eye out: with camera or screen shared she may speak up on her own (you come back, you look upset, you hold something up), never while you talk
• Show camera / Show screen: one still goes with every message so she reacts to you and your screen
• Lighter on the GPU: the image engine unloads when idle, her eyes unload after every look, her brain reserves only the context she uses

Agent Gen 0.5.34

Unity, live, in one window. Her face moves when she talks, her voice is her own, she hears you, and she becomes whatever you tell her to be.

Download AgentGen-Setup-0.5.34.exe below, run it, pick a drive with room. The first launch asks your age (18+ only), then downloads what she runs on (about 80 GB: Python, PyTorch, her models, her brain).

Needs: Windows 11 64-bit, an NVIDIA GPU with 16 GB VRAM (RTX 30/40/50), 32 GB RAM, 100 GB of free storage for the install folder, speakers and a mic. 18+.

Privacy: nothing you say or show leaves your machine. The one exception is her new web search: the search words and the pages she opens go to DuckDuckGo/Bing and those sites, and every one is listed in the chat. WEB=0 turns it off. See PRIVACY.md.

The installer is not code-signed yet, so Windows SmartScreen may warn: More info → Run anyway.

In this release (0.5.32 – 0.5.34)

• Her lips stay on her words: each sentence waits until enough of her face is rendered to play in step with her voice (her face renders below 25 fps, so her lips used to fall behind)
• Age gate at first launch: confirm 18+ and your birth date, or the app closes
• She looks things up: ask about news, scores, weather or a page and she searches the web, reads it and tells you; links show in the chat
• She keeps an eye out: with camera or screen shared she may speak up on her own (you come back, you look upset, you hold something up), never while you talk
• Show camera / Show screen: one still goes with every message so she reacts to you and your screen
• Lighter on the GPU: the image engine unloads when idle, her eyes unload after every look, her brain reserves only the context she uses

Agent Gen 0.5.31

Unity, live, in one window. Her face moves when she talks, her voice is her own, she hears you, and she becomes whatever you tell her to be.

Download AgentGen-Setup-0.5.31.exe below, run it, pick a drive with room. The first launch downloads what she runs on (about 80 GB: Python, PyTorch, her models, her brain), then she needs no internet at all.

Needs: Windows 11 64-bit, an NVIDIA GPU with 16 GB VRAM (RTX 30/40/50), 32 GB RAM, 100 GB of free storage for the install folder, speakers and a mic. 18+.

Privacy: nothing leaves your machine. Everything runs over 127.0.0.1, with no telemetry, no cloud and no update checks. See PRIVACY.md.

License: MIT for Agent Gen itself. The models and runtimes it installs keep their own licenses; see THIRD-PARTY-NOTICES.md (the new clothes-parsing model is non-commercial).

The installer is not code-signed yet, so Windows SmartScreen may warn on first run: More info → Run anyway.

In this release (0.5.25 – 0.5.31)

• She keeps her look: taking clothes off or adding something edits only her body below the chin, so her face, hair and room stay hers, and she no longer strips or swaps outfits on her own
• Her look survives a restart; Reset puts her default picture back on screen
• Write or draw on her in marker, paint or lipstick ("write BRAT on your right thigh in purple paint", "a heart on your chest in lipstick"): it lands on exactly that part of her, skin or clothes, sized to fit, readable
• Shut down button: unloads her from the GPU, stops every Agent Gen process and closes the app
• Fixed: "write ... across your skin" no longer turns her into a different woman; marker, paint and lipstick always keep her
• Marker, paint and lipstick are drawn from your words even when she answers with a movement; "write "hole" with a down arrow" draws both, the arrow under the word
• The setup wizard states what the folder really needs: 100 GB free, NVIDIA 16 GB GPU, 32 GB RAM, Windows 11

Agent Gen 0.5.30

Unity, live, in one window. Her face moves when she talks, her voice is her own, she hears you, and she becomes whatever you tell her to be.

Download AgentGen-Setup-0.5.30.exe below, run it, pick a drive with room. The first launch downloads what she runs on (about 80 GB: Python, PyTorch, her models, her brain), then she needs no internet at all.

Needs: Windows 11 64-bit, an NVIDIA GPU with 16 GB VRAM (RTX 30/40/50), 32 GB RAM, about 120 GB free disk, speakers and a mic. 18+.

Privacy: nothing leaves your machine. Everything runs over 127.0.0.1, with no telemetry, no cloud and no update checks. See PRIVACY.md.

License: MIT for Agent Gen itself. The models and runtimes it installs keep their own licenses; see THIRD-PARTY-NOTICES.md (the new clothes-parsing model is non-commercial).

The installer is not code-signed yet, so Windows SmartScreen may warn on first run: More info → Run anyway.

In this release (0.5.25 – 0.5.30)

• She keeps her look: taking clothes off or adding something edits only her body below the chin, so her face, hair and room stay hers, and she no longer strips or swaps outfits on her own
• Her look survives a restart; Reset puts her default picture back on screen
• Write or draw on her in marker, paint or lipstick ("write BRAT on your right thigh in purple paint", "a heart on your chest in lipstick"): it lands on exactly that part of her, skin or clothes, sized to fit, readable
• Shut down button: unloads her from the GPU, stops every Agent Gen process and closes the app
• Fixed: "write ... across your skin" no longer turns her into a different woman; marker, paint and lipstick always keep her

Agent Gen 0.5.29

Unity, live, in one window. Her face moves when she talks, her voice is her own, she hears you, and she becomes whatever you tell her to be.

Download AgentGen-Setup-0.5.29.exe below, run it, pick a drive with room. The first launch downloads what she runs on (about 80 GB: Python, PyTorch, her models, her brain), then she needs no internet at all.

Needs: Windows 11 64-bit, an NVIDIA GPU with 16 GB VRAM (RTX 30/40/50), 32 GB RAM, about 120 GB free disk, speakers and a mic. 18+.

Privacy: nothing leaves your machine. Everything runs over 127.0.0.1, with no telemetry, no cloud and no update checks. See PRIVACY.md.

License: MIT for Agent Gen itself. The models and runtimes it installs keep their own licenses; see THIRD-PARTY-NOTICES.md (the new clothes-parsing model is non-commercial).

The installer is not code-signed yet, so Windows SmartScreen may warn on first run: More info → Run anyway.

In this release (0.5.25 – 0.5.29)

• She keeps her look: taking clothes off or adding something edits only her body below the chin, so her face, hair and room stay hers, and she no longer strips or swaps outfits on her own
• Her look survives a restart; Reset puts her default picture back on screen
• Write or draw on her in marker, paint or lipstick ("write BRAT on your right thigh in purple paint", "a heart on your chest in lipstick"): it lands on exactly that part of her, skin or clothes, sized to fit, readable
• Shut down button: unloads her from the GPU, stops every Agent Gen process and closes the app

Agent Gen 0.5.24

Unity, live, in one window. Her face moves when she talks, her voice is her own, she hears you, and she becomes whatever you tell her to be.

Download AgentGen-Setup-0.5.24.exe below, run it, pick a drive with room. The first launch downloads what she runs on (about 80 GB: Python, PyTorch, her models, her brain), then she needs no internet at all.

Needs: Windows 11 64-bit, an NVIDIA GPU with 16 GB VRAM (RTX 30/40/50), 32 GB RAM, about 120 GB free disk, speakers and a mic. 18+.

Privacy: nothing leaves your machine. Everything runs over 127.0.0.1, with no telemetry, no cloud and no update checks. See PRIVACY.md.

License: MIT for Agent Gen itself. The models and runtimes it installs keep their own licenses; see THIRD-PARTY-NOTICES.md.

The installer is not code-signed yet, so Windows SmartScreen may warn on first run: More info → Run anyway.

In this release

• Outfit and clothing changes keep the same girl: her current picture is edited, not redrawn
• Talking to her mid-render never jams: the video engine is aborted and restarted if it will not stop
• She moves her whole body (Wan 2.2 at 480x832), on her own when idle, and loops instead of freezing while the next clip renders
• Looks render fresh from her words; tell her she's a man, a lion, a gorilla, and she is; "that's you now" adopts a picture
• [[pic]] pictures of anything, in the chat
• Storage panel: see and clear her looks, clips, logs and conversations, or reset to fresh without reinstalling
• Installer stops everything it finds running (orphaned brain runners included) and never shows "cannot be closed"
• Talking to her always wins: renders are dropped so she answers right away

Agent Gen 0.5.23

Unity, live, in one window. Her face moves when she talks, her voice is her own, she hears you, and she becomes whatever you tell her to be.

Download AgentGen-Setup-0.5.23.exe below, run it, pick a drive with room. The first launch downloads what she runs on (about 80 GB: Python, PyTorch, her models, her brain), then she needs no internet at all.

Needs: Windows 11 64-bit, an NVIDIA GPU with 16 GB VRAM (RTX 30/40/50), 32 GB RAM, about 120 GB free disk, speakers and a mic. 18+.

Privacy: nothing leaves your machine. Everything runs over 127.0.0.1, with no telemetry, no cloud and no update checks. See PRIVACY.md.

License: MIT for Agent Gen itself. The models and runtimes it installs keep their own licenses; see THIRD-PARTY-NOTICES.md.

The installer is not code-signed yet, so Windows SmartScreen may warn on first run: More info → Run anyway.

In this release

• Talking to her mid-render never jams: the video engine is aborted and restarted if it will not stop
• She moves her whole body (Wan 2.2 at 480x832), on her own when idle, and loops instead of freezing while the next clip renders
• Looks render fresh from her words; tell her she's a man, a lion, a gorilla, and she is; "that's you now" adopts a picture
• [[pic]] pictures of anything, in the chat
• Storage panel: see and clear her looks, clips, logs and conversations, or reset to fresh without reinstalling
• Installer stops everything it finds running (orphaned brain runners included) and never shows "cannot be closed"
• Talking to her always wins: renders are dropped so she answers right away

Agent Gen 0.5.22

Unity, live, in one window. Her face moves when she talks, her voice is her own, she hears you, and she becomes whatever you tell her to be.

Download AgentGen-Setup-0.5.22.exe below, run it, pick a drive with room. The first launch downloads what she runs on (about 80 GB: Python, PyTorch, her models, her brain), then she needs no internet at all.

Needs: Windows 11 64-bit, an NVIDIA GPU with 16 GB VRAM (RTX 30/40/50), 32 GB RAM, about 120 GB free disk, speakers and a mic. 18+.

Privacy: nothing leaves your machine. Everything runs over 127.0.0.1, with no telemetry, no cloud and no update checks. See PRIVACY.md.

License: MIT for Agent Gen itself. The models and runtimes it installs keep their own licenses; see THIRD-PARTY-NOTICES.md.

The installer is not code-signed yet, so Windows SmartScreen may warn on first run: More info → Run anyway.

In this release

• She moves her whole body (Wan 2.2 at 480x832), on her own when idle, and loops instead of freezing while the next clip renders
• Looks render fresh from her words; tell her she's a man, a lion, a gorilla, and she is; "that's you now" adopts a picture
• [[pic]] pictures of anything, in the chat
• Storage panel: see and clear her looks, clips, logs and conversations, or reset to fresh without reinstalling
• Installer stops everything it finds running (orphaned brain runners included) and never shows "cannot be closed"
• Talking to her always wins: renders are dropped so she answers right away