EchoBot

Audience

People exploring the SideQuest site and builders interested in interactive voice characters.

The challenge

Connect a useful conversation to a character whose face and speech reflect the actual runtime state.

The approach

Give Rive, reply generation, speech, and structured decisions clear roles. Verify the native character states through Rive CLI and check the complete interaction in the browser.

What SideQuest Studio handled

Character and interaction design · Rive scene development · Local AI and speech integration · Streaming coordination · Live deployment verification

Important decisions

• Use a mouthless idle face and a separate speaking channel.
• Consume generated reply text without waiting for Laya’s classification.
• Keep temporary voice impressions separate from normal conversation memory.
• Use accessible HTML controls alongside the Rive character.

Testing and iteration

• Verify and inspect the Rive project; render idle, speaking, thinking, and gesture states.
• Check live replies and valid speech audio.
• Test a delayed Laya decision and the return from an impression to normal delivery.
• Check mobile layout and reduced motion.

Current outcome

A deployed website companion with a native Rive face, local replies and speech, and a clearer contract between conversation and character state.

Known limitations

• Reply and story quality vary with the small local model.
• Cold starts, inference, speech generation, and network delivery still add latency.

Next steps

• Evaluate more real conversations and short stories.
• Continue refining character reactions against actual playback.