QWE AI Academy End state: a Unity scene with an NPC that hears you through your mic, replies in a synthesized voice with matching lip-sync, and remembers what you told it last session. All the game NPC AI plumbing – speech-to-text, LLM reasoning, TTS, and animation – runs through Convai’s cloud. Your Unity project just holds the components.
This guide walks the deploy path for Convai Unity SDK v4.2.0 – the current release as of mid-2026 – plus the three install errors that eat afternoons and the pricing trap that catches almost every new developer.
What you’re actually deploying
Convai isn’t a local model. It’s an SDK that pipes voice input from your game to Convai’s servers, runs it through STT + LLM + TTS, and streams audio back with lip-sync data. According to NVIDIA’s developer blog, Convai integrates NVIDIA Avatar Cloud Engine – using NVIDIA Audio2Face for facial animation and NVIDIA Riva for speech-to-text and text-to-speech. On the lip-sync side, Convai’s own Unity tutorial describes NeuroSync: a gen-AI lip-sync model that analyzes audio in real-time to drive facial animations compatible with ARKit, CC4, and MetaHuman, streamed via WebRTC.
Translation: your NPC needs an internet connection to talk. There’s no offline mode, and every conversation counts against a monthly quota.
System requirements
The Unity plugin has a narrow but forgiving compatibility window. The Unreal SDK is a separate track – this guide covers Unity, with a note on Unreal at the end.
| Requirement | Details |
|---|---|
| Unity Editor | 2022.3.x minimum (per official compatibility docs as of mid-2026) |
| OS | Check docs.convai.com/compatibility – platform support may have changed |
| Dependencies | Newtonsoft.Json, Animation Rigging, Input System – UPM auto-installs all three |
| Network | Stable broadband required; low-latency connection matters (see latency gotcha in FAQ) |
| Convai account | Free tier (100 interactions/mo) or Indie Dev ($22-29/mo as of mid-2026) |
One thing the docs bury: the SDK depends on gRPC native binaries. On macOS you’ll hit a security-permission prompt for grpc_csharp_ext.bundle the first time you press Play. Approve it in System Settings → Privacy & Security, or the build will silently fail to reach the Convai backend.
Get an API key first
Skip this and every setup step below is wasted work.
- Go to convai.com and create an account.
- Land on the Dashboard. There’s an API key at the top – copy it.
- Optional but recommended: create a test character in the Playground now. You’ll get a Character ID string, which the demo scene needs.
The free Creator tier gives you enough to verify install works but nothing more. That’s 100 interactions per month, a 1MB knowledge bank, and a 50-interaction cap on flagship LLMs (as of mid-2026, per Convai’s pricing page). Fine for the demo scene. Not fine for a jam.
Install Convai Unity SDK v4.2.0
UPM is the path Convai officially recommends. The Asset Store zip works too, but UPM is easier to maintain and keep consistent across a team – both methods install the identical package according to Convai’s installation docs. Skip the zip unless your studio pins packages through My Assets.
# In Unity Editor:
# 1. Window > Package Manager
# 2. Click the + button (top-left)
# 3. Select "Add package by name"
# 4. Paste the Convai package name from docs.convai.com
# 5. Click Install
# Unity will auto-resolve three dependencies:
# - com.unity.inputsystem
# - com.unity.ugui
# - Newtonsoft.JsonWait for the progress bar in the bottom-right to finish. Installation is complete when the Convai SDK for Unity entry appears in the Package Manager list with version 4.2.0 and a new Convai menu item appears in the Unity menu bar.
Now configure the key. Top menu → Convai → Account → paste your API key → Save API Key. Your usage data should update on the dashboard to confirm the connection. If usage stays at zero, the key didn’t take – regenerate it from the dashboard.
Verify it works in under 5 minutes
The plugin ships with a full-features demo scene. Use it to prove the pipeline is live before you touch your own characters.
- In your Hierarchy, click the + icon → Convai → Setup Required Components. This drops the Convai Manager into your scene.
- Open the demo scene:
Convai > Demo > Scenes > Full Features. - Select the Amelia NPC. In the Inspector, either use the default Character ID or paste one from your Playground.
- Press Play. Walk close to Amelia. Hold T and speak.
Expected result: the mic icon lights up, your words appear as a transcript, Amelia responds with voice + lip-sync. If audio comes back but there’s no transcript, that’s a mic-permission issue on the OS layer, not the plugin.
Pro tip: Before shipping anything, check the developer forum at forum.convai.com for known regressions in your exact SDK version. Convai’s Unreal SDK 3.6.9-hotfix-2 broke long-term memory – affected users reported characters claiming no memory of prior conversations, directly contradicting what paid plans advertise (documented in a March 2026 community review). Backend regressions like this don’t always make the changelog.
The three errors that actually happen
Not theoretical. These are the ones the forum and docs troubleshooting page cover because they hit almost every install.
1. DllNotFoundException: grpc_csharp_ext
Symptom: character loads in the demo scene, then Play throws DllNotFoundException: grpc_csharp_ext and nothing responds. This exact error is documented in Convai Developer Forum post #2630 – the character loads normally and then the gRPC native extension fails.
Cause: the gRPC native library got quarantined (macOS) or Assembly Validation is checking exact version numbers (Windows). Fix on macOS: right-click the .bundle in Finder → Open → allow. Fix on Windows: see error #2 below.
2. Assembly reference mismatch (System.Memory 4.0.1.1 vs 4.0.1.2)
The console throws: Grpc.Core.Api references strong named System.Memory Assembly references: 4.0.1.1 Found in project: 4.0.1.2. The fix: in Project Settings → Player → Other Settings, make sure Assembly Validation is disabled, then restart Unity. That’s it. The underlying reason – Unity checks exact version numbers by default for included assemblies, but this plugin uses the latest libraries and doesn’t need that strictness – is explained in Convai’s troubleshooting docs.
3. Dependency conflict on install
If your project already pins com.unity.inputsystem or com.unity.ugui to an older version in Packages/manifest.json, the install will fail or produce a version mismatch. Open manifest.json, remove or update the conflicting version entries, then retry. This one hits studios more than solo devs because studios often lock package versions across projects.
The pricing gotcha nobody mentions in tutorials
Convai advertises plans by “interactions per month.” But there’s a second, silent cap.
Per Convai’s own FAQ (as of mid-2026), each plan has a Flagship LLM Interaction Cap – the limit on interactions using flagship models like GPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro. On the Indie Dev plan you get 3,000 total monthly interactions, but the flagship cap is 1,500. Once you hit 1,500 interactions on GPT-4o, your flagship quota is gone and the remaining 1,500 auto-downgrade to a non-flagship LLM – without warning.
Practically: your NPC gets noticeably dumber halfway through the month, and if you don’t know why, you’ll blame your prompt.
Turns out the concurrency limit works the same way – per API key, not per NPC. With a concurrency limit of 1, you can have 10 AI NPCs in your scene, but only one user can speak to any of them at the same time. For multiplayer, that’s not a code problem. It’s a pricing tier decision you need to make before you build.
Upgrade, downgrade, uninstall
Upgrading is boring, which is a feature. Return to My Assets in the Package Manager, select the SDK, and click Update. Same for UPM – the package resolves through Convai’s registry.
Uninstall: Package Manager → select Convai SDK for Unity → Remove. Then manually delete the Assets/Convai folder and the Resources/APIKey asset if either persists. For a clean slate, also nuke the Library folder and let Unity rebuild.
For Unreal developers, the equivalent path lives on GitHub: the Convai-UnrealEngine-SDK-V4 releases page. Per Convai’s Unreal docs, engine-level installation is the recommended method if you want the plugin available across all projects on the same engine version.
FAQ
Does Convai work offline?
No. Every voice interaction hits Convai’s cloud for STT, LLM, and TTS. No local inference mode exists.
Why is my NPC taking 8+ seconds to reply?
Not your code. Developers in Convai’s own forum reported 8-10 second response times across all characters in March 2026 – one user noted the STT service was also cutting out, and the pattern was consistent enough to qualify as a recurring backend issue, not a one-off. Check status.convai.com and the forum before assuming your project is broken. If latency is consistent throughout the day, your options are limited: switch the character to a lighter LLM in the Playground, or accept that response time is a backend variable you can’t fix locally.
Can I ship a commercial game on the Indie Dev plan?
Technically yes, but budget for the Professional tier before launch. Between the 1,500 flagship-LLM cap and the concurrency limit (typically 1 on Indie Dev), a public build will exhaust your quota in hours, not weeks. The free tier is fine for prototyping; Indie Dev at $22/month annual is fine for closed alpha. Anything with concurrent public users needs Professional or Enterprise.
Next step: open the Playground, upload a 2-3 page backstory PDF to your character’s knowledge bank, and test whether the character actually references it in conversation. That’s the one capability that separates Convai from a plain LLM wrapper – and the fastest way to know if the platform fits your project.