Editor's note (8 October 2026): Eposyne Director is built to run on-device through its native SDK, which is not yet available. Today, developers use the hosted Director beta. Director has plugins for Unity and Unreal, not Godot, and its story-arc module for plot progression is in development. We have corrected the table, the engine lists, and the FAQ; where the article still describes Director as on-device, that describes the native SDK.
Editor's note (14 October 2026): The hosted Director beta's model calls are paid from usage credits, and every new account starts with a welcome credit to try it. Director's licence has no fee until your game launches, then takes 1% of gross revenue above the first EUR 100,000. We have corrected the lines below that called Director free or free of per-token fees today.
Inworld AI and Convai are the two names most developers reach for when they want AI-driven NPCs. Both are capable, and both are worth outgrowing for the same reasons: they bill per conversation, they run in the cloud, and one of them has quietly changed direction. If you are shopping for an alternative, the useful question is not "what is like Convai" but "which layer of the problem am I actually solving, and do I want it in the cloud or on the player's machine."
This guide lays out the real alternatives in 2026, groups them by what they actually do, and helps you choose. It includes our own runtime, Eposyne Director, and points you elsewhere when another tool fits better.
Two things to know before you shop:
- Inworld has shifted away from being a games-first character studio toward a general real-time AI runtime and voice stack. It still works for games, but it is no longer built only for you, and that alone sends a lot of teams looking.
- "AI NPC" and "AI game master" are different products. Most tools here make individual characters talk. A smaller group, which is where Eposyne Director sits, runs the world: pacing, events, and how the whole story reacts. Know which one you need. We cover that distinction in detail in on-device AI game master middleware.
What to evaluate in an alternative
Five axes separate these tools:
- Cloud vs on-device. Cloud is easy to start and bills per conversation for as long as the servers answer. Running the model on the player's hardware removes that bill and the network dependency at the cost of some GPU budget, which is the trade GladeCore, NVIDIA ACE, and Eposyne Director each make in different parts of the stack.
- Layer: character vs director. Convai, GladeCore, and Charisma.ai give an individual NPC a voice and memory. Eposyne Director works a layer above that, deciding which events fire, how pacing moves, and how the world reacts to what the player actually did.
- Cost model. Usage-based costs scale with your players. Fixed costs do not.
- Engine support. Unreal, Unity, Godot, custom.
- Control. How hard is it to keep the AI in character and inside your lore.
The alternatives at a glance
| Tool | What it is | On-device | Cost model | Engines |
|---|---|---|---|---|
| Convai | Cloud NPC characters with perception and voice | No | Usage-based | Unreal, Unity |
| Inworld | Cloud runtime and voice, pivoted from NPC studio | No | Usage-based | Unreal, Unity |
| GladeCore | On-device NPC dialogue | Yes | Vendor tool | Unreal, Unity |
| NVIDIA ACE | On-device character suite on RTX (speech, SLM, animation) | Yes (RTX) | Vendor SDK | Unreal, custom |
| Charisma.ai | Narrative-constrained conversational characters | No (cloud) | Usage-based | Unreal, Unity |
| Eposyne Director | Game master, constraint-validated | Not yet (native SDK) | Hosted beta from usage credits; licence free until launch, then 1% above first EUR 100k | Unreal, Unity, others through the API |
| Roll your own | Local LLM plus your own orchestration | Yes | Your infrastructure | Any |
Pricing and hardware requirements move fast in this space. Check each vendor's current documentation before you commit.
The alternatives in detail
On-device NPC dialogue: GladeCore
If what you need is talking characters without the cloud bill, GladeCore is the closest like-for-like swap for Convai. It runs speech, a local model, and voice on the player's machine inside Unreal and Unity, so there is no per-conversation fee after you ship. It is a character-layer tool, not a director. For a game that needs a handful of NPCs that hold a conversation, and a team that wants to escape usage pricing, it is a strong pick.
High-fidelity characters on RTX: NVIDIA ACE
NVIDIA ACE is a suite for AI characters that can run on-device on RTX hardware, spanning speech recognition, small language models, and facial animation, with an agent layer that can reach toward director-like behavior. It is the most production-hardened on-device option and the most demanding, targeting RTX-class GPUs and real engineering investment. Best for AAA or high-fidelity projects already on NVIDIA's stack.
Narrative-first characters: Charisma.ai
If your NPCs must stay on-story, Charisma is built for authored, branching narrative with characters that stay inside the writer's boundaries. It runs in the cloud and is less suited to open-world improvisation, but for story-driven games where dialogue must not wander, it is a strong option.
The director layer: Eposyne Director
The tools above make characters talk. Director runs the world. It reads your game state and returns constraint-validated narrative decisions: NPC behavior, plot progression, and emergent events, all checked against rules you define so the model cannot break your lore. It has plugins for Unreal and Unity (other engines can call its API), pays for its hosted beta's model calls from usage credits, and has no licence fee until your game launches, then takes 1% of gross revenue above the first EUR 100,000. Honest status: Director is in beta, and its authoring companion, Architect, is free to start. Best if you want systemic narrative across a whole game rather than a single chatty NPC. For a direct head-to-head, see Eposyne vs Inworld and Eposyne vs Convai.
Build it yourself
Many shipped games skip platforms entirely: a small local model with a runner like llama.cpp, Ollama, or Unity Sentis, wrapped in your own memory, retrieval, and validation. Maximum control, no vendor lock-in, and the most work. If you have the engineering time, it is the most flexible route. We walk through the whole build in how to run a local LLM for game NPCs. Eposyne Director is that same on-device architecture sold as middleware, so the comparison worth doing is your engineering months against Director's 1% of gross revenue above the first EUR 100,000.
Which should you choose?
- On-device talking characters, escape the cloud bill: GladeCore.
- AAA fidelity on RTX with animation: NVIDIA ACE.
- Authored, on-story characters: Charisma.ai.
- A director that drives systemic narrative on-device and enforces your rules: Eposyne Director (beta).
- Total control, and you have the engineers: roll your own.
- Fast prototype and cloud is fine: honestly, Convai is still good at this, and Inworld if you want its voice stack.
The layers combine. Many teams pair a director for orchestration with a character tool for a few hero NPCs.
FAQ
Is Inworld AI still good for games? It still works, and its voice models are strong, but Inworld has repositioned toward a general real-time AI runtime rather than a games-first NPC studio. If you specifically want a game-character platform, that shift is worth weighing.
What is the cheapest alternative to Convai? On-device tools remove the per-conversation fee entirely, so they tend to win on long-run cost once you ship, at the price of using the player's hardware. Among them, GladeCore charges a vendor fee for the character layer, and Eposyne Director's native SDK, which is not yet available, would run on the player's hardware under a licence that is free until your game has earned EUR 100,000 gross and then takes 1% above that floor; today's hosted Director beta is paid per call from usage credits. Cloud tools are cheaper to start and more expensive to scale with players.
Are there on-device alternatives to Inworld and Convai? Yes. GladeCore runs NPC dialogue locally, NVIDIA ACE runs on RTX hardware, Eposyne Director is built to run the game-master layer on-device (its native SDK is not yet available), and you can build your own with a local LLM.
What is the difference between an AI NPC and an AI game master? An AI NPC is one character that talks and remembers. An AI game master orchestrates the whole experience: events, pacing, and how the world reacts. Some games want one, some want both.
Which engines do these tools support? Most support Unreal and Unity; some add Godot and custom engines. Director has plugins for Unreal and Unity, and other engines can call its API.