Dynamic Context
Learn how Dynamic Context enables more relevant, adaptive, and context-aware AI character interactions.
Introduction
Convai characters have two layers of context:
Character Level — backstory, personality, and behavioral rules. Configured in the Convai dashboard, persists across all sessions.
Session Level — information relevant only to the current interaction: who the player is, what just happened, the current game state.
Dynamic Context is a runtime feature of the Convai Unity SDK that lets your game push ephemeral, session-specific information to a Convai AI character during an active conversation. Unlike a character's fixed backstory configured in the Convai dashboard, Dynamic Context is injected at runtime and is temporary — it exists only for the duration of the current session.
Use it to bridge the gap between your game world and the AI: tell the character what just happened, who the player is, what they're carrying, where they are, or anything else that should influence the character's responses without being hardcoded into a prompt.
How It Works
When you call UpdateDynamicContext, the SDK serializes a context-update message and delivers it to the Convai backend over the active session's data channel. The backend receives the update and folds the new information into the character's active context window before processing the next user utterance.
The flow is:
Your Game Code
│
▼
IConvaiRoomConnectionService.UpdateDynamicContext(text, mode, runLlm)
│
▼
Serialized context-update message dispatched over the session data channel
│
▼
Convai Backend → LLM Context WindowThis is a fire-and-forget, outbound-only operation. There is no direct response to the context update itself — the effect is visible in the character's next spoken reply.
Key Concepts
Before writing any code, understand these three parameters that control every UpdateDynamicContext call.
text
textThe content you want the character to be aware of. This is a plain string — it can be a sentence, a structured paragraph, or key-value data formatted however you prefer.
text is required for append and replace modes. When using reset mode, it is ignored.
mode
modeControls how the new text is applied to the character's existing ephemeral context.
append
Adds text to any existing ephemeral context. Prior entries remain.
Progressive information — adding facts as the session unfolds.
replace
Discards all current ephemeral context and replaces it entirely with the new text.
A clean state transition — entering a new scene, starting a new chapter.
reset
Clears all runtime ephemeral context. text is ignored. Initial Dynamic Info is not affected — it remains pinned for the duration of the session.
Resetting accumulated runtime context while preserving session-level fixed context.
Default: "append"
runLlm
runLlmWhether to trigger an LLM inference pass after the context update.
auto
The server decides based on its internal pipeline state. Recommended for most use cases.
true
Always triggers an LLM response immediately after the context update.
false
Updates the context silently — no LLM response is triggered.
Default: "auto"
Use "false" when you are batching multiple context updates and only want the LLM to respond after the final one. Use "true" when the context update itself should prompt a reaction from the character.
mode
modeControls how the new text is applied to the character's existing ephemeral context.
append
Adds text to any existing ephemeral context. Prior entries remain.
Progressive information — adding facts as the session unfolds.
replace
Discards all current ephemeral context and replaces it entirely with the new text.
A clean state transition — entering a new scene, starting a new chapter.
reset
Clears all runtime ephemeral context. text is ignored. Initial Dynamic Info is not affected — it remains pinned for the duration of the session.
Resetting accumulated runtime context while preserving session-level fixed context.
Default: "append"
Initial Dynamic Info
UpdateDynamicContext is not the only way to provide session-level context to a character. The ConvaiCharacter component exposes a second layer that is fixed for the entire session: Initial Dynamic Info.
This is configured in the Unity Inspector on the ConvaiCharacter component, under the SESSION section:
Enable Keep Initial Dynamic Info In Context — this toggle controls whether the text below is included in the connection request and pinned by the server for the whole session.
Once enabled, the Dynamic Info (Connection Request) field appears. Enter your initial context text in Initial Dynamic Info Text.

Unlike UpdateDynamicContext, this value is not sent over the data channel at runtime — it is embedded in the connection payload before the session starts.
Key distinction: reset mode in UpdateDynamicContext does not clear Initial Dynamic Info. It only resets the runtime ephemeral layer. The pinned initial text remains active for the entire session regardless of any context updates you send.
When to use each
Set via
Unity Inspector (ConvaiCharacter)
Code (UpdateDynamicContext)
When sent
At room connection, before session starts
Any time during an active session
Mutable
No — fixed for the session
Yes — append, replace, or reset at will
Cleared by reset
No
Yes
Use for
Stable session configuration: character role, scenario rules, fixed instructions
Changing game state, player data, events, mood
Use Initial Dynamic Info for context that should never change during a session — role instructions, scenario setup, fixed rules. Use UpdateDynamicContext for everything that evolves as the session progresses.
Accessing the API
Dynamic Context is exposed through IConvaiRoomConnectionService, obtained via ConvaiManager. Two distinct patterns exist depending on who owns the connection lifecycle.
Pattern A — Self-Contained
Use when the component is responsible for its own initialization. Typical for session-start scenarios where context must be sent as soon as the connection is ready.
Pattern B — Inject
Use when a session coordinator or parent system is responsible for providing a connected service. The component assumes the service is valid when its public methods are called.
UpdateDynamicContext returns false if the session is not connected. Always check the return value.
API Reference
text
string
Yes (unless reset)
—
The context text to inject.
mode
string
No
"append"
How to apply the text: "append", "replace", or "reset".
runLlm
string
No
"auto"
Whether to trigger an LLM response: "true", "false", "auto".
Returns: true if the message was dispatched to the transport layer; false if the session is not connected.
mode
modeControls how the new text is applied to the character's existing ephemeral context.
append
Adds text to any existing ephemeral context. Prior entries remain.
Progressive information — adding facts as the session unfolds.
replace
Discards all current ephemeral context and replaces it entirely with the new text.
A clean state transition — entering a new scene, starting a new chapter.
reset
Clears all runtime ephemeral context. text is ignored. Initial Dynamic Info is not affected — it remains pinned for the duration of the session.
Resetting accumulated runtime context while preserving session-level fixed context.
Default: "append"
Dynamic Context vs. Related Features
Dynamic Context
UpdateDynamicContext
Ephemeral (runtime)
No
Runtime facts: player state, game events, situational awareness.
Initial Dynamic Info
ConvaiCharacter Inspector
Ephemeral (session-start, pinned)
No
Fixed session instructions: scenario rules, character role, stable setup. Not affected by reset.
Dynamic Info
SendDynamicInfo
Per-message
No
Inline context attached to a specific user utterance.
Template Keys
UpdateTemplateKeys
Session
No
Resolving named placeholders ({PlayerName}) in narrative design text.
Scene Metadata
UpdateSceneMetadata
Session
No
Structural scene data for the backend's contextual grounding pipeline.
Dashboard Backstory
Convai dashboard
Permanent
Yes
Character personality, lore, fixed behavioral rules.
Common Pitfalls
Calling before the session is connected
UpdateDynamicContextreturnsfalseand silently drops the update. Use a coroutine or subscribe toIConvaiRoomConnectionService.Connectedto delay context injection until the session is live.Using
appendindefinitely Everyappendexpands the context window. In long sessions this degrades response coherence and increases latency. Usereplaceperiodically, or adopt theDynamicContextManagerpattern.Sending
textwithresetmodetextis ignored whenmodeis"reset". Passnullexplicitly to make the intent clear.Expecting a synchronous effect Dynamic Context is asynchronous end-to-end. The character will not know about the update until the next LLM inference pass. Do not design logic that assumes an instant effect.
Overusing
runLlm: "true"Forcing an immediate response on every update creates conversational noise. Reserve it for moments where a character reaction is genuinely expected.
Quick Reference
Conclusion
Dynamic Context is the bridge between your game or application's live state and your AI character's understanding of it. The character's permanent configuration — personality, backstory, behavioral rules — stays stable in the Convai dashboard. Everything that belongs to this session, this user, and this moment flows through UpdateDynamicContext.
The API is intentionally simple: a single method with three parameters. The sophistication lives in how you use it — whether that is personalizing a conversation at session start, reacting to a threshold crossing mid-session, silently accumulating events until they are worth commenting on, or tracking competency gaps across an entire training session.
A few principles to carry forward:
Silent updates are the default posture. Use
runLlm: "false"for state synchronization. ReserverunLlm: "true"for moments that genuinely warrant a character response.Treat the context window as a resource. Unbounded
appenddegrades quality over time. Usereplaceat natural boundaries — scene transitions, module changes, session resets.Your application owns the state. The character receives it. Dynamic Context is the mechanism that enforces that separation cleanly.
Need help? For questions, please visit the Convai Developer Forum.
Last updated
Was this helpful?