Pipecat 1.15.0, SkillSpector 2.13.3, Laya Guard 0.7.1
Updated 2 min read
Summary
Agent Radar 2026-11-03: Pipecat 1.15.0 adds expressive TTS and turn modes; SkillSpector 2.13.3 raises scan limits; Laya Guard 0.7 fine-tunes its model.
Agent Radar tracked three launches as of 2026-11-03, covering the window 2026-10-27 to 2026-11-03. One is a voice agent framework and two are safety tools for coding agents: one scans skills before install, the other judges each command at run time. All three change defaults, so check the upgrade notes first.
At a glance
| Launch | Tool | Released | What changed |
|---|---|---|---|
| Pipecat 1.15.0 adds expressive TTS and turn strategy modes | Pipecat | 2026-10-29 | Alpha TTS expression tags, Pipecat Tone models, classifier turn ending, AssemblyAI TTS; multiple stop strategies must now all agree |
| SkillSpector 2.13.3 raises scan limits, checks AI replies | SkillSpector | 2026-11-02 | Configurable TP4 batch cap and 18-minute deadline, refused or truncated AI responses reported as incomplete, scan-integrity fixes |
| Laya Guard 0.7 adds a fine-tuned model and rewrite blocks | Laya Guard | 2026-11-03 | Fine-tuned Laya model (~650 MB), rewritten rm -rf forms blocked, credential reads and on-chain transactions handled by rules |
What changed
- Pipecat: v1.15.0 lets the LLM write delivery and sound tags that Cartesia, ElevenLabs, Fish Audio, MiniMax and five other TTS services perform. The behaviour change is in turn-taking:
stop_modenow defaults toStrategyMode.ALL. - SkillSpector: 2.13.3 lets large skills finish the TP4 check and no longer treats an unfinished AI review as clean. Regenerate version-bound baselines after upgrading.
- Laya Guard: 0.7.0 and 0.7.1 swap in a fine-tuned model that allows more everyday commands and holds more attacks on the project's own test sets.
cast wallet decrypt-keystoreandgh auth tokenare now treated as credential reads.
Questions
Which item needs action first?
Pipecat 1.15.0, if your bot has more than one stop strategy or a wake phrase, because turns can end later. Set stop_mode=StrategyMode.FIRST to keep the old timing.
Which item is safe to take without changes?
Laya Guard 0.7.1 for most users, apart from the one-time model download. Set LAYA_GUARD_MODEL=base to keep the old model.
New guide today: Pipecat vs LiveKit Agents. Related reading: best free AI voice models.
In this issue
Rank 1 Pipecat 1.15.0 adds expressive TTS and turn strategy modes
PipecatPipecat 1.15.0 (Oct 29, 2026) adds alpha expressive TTS tags, classifier turn ending and AssemblyAI TTS; multiple stop strategies now must agree.
Category: Voice agents
Pricing: Open source
Rank 2 SkillSpector 2.13.3 raises scan limits, checks AI replies
SkillSpectorNVIDIA SkillSpector 2.13.3 (Nov 2, 2026) makes TP4 batch and deadline limits configurable and reports refused or truncated AI reviews as incomplete.
Category: Agent infra
Pricing: Open source
Rank 3 Laya Guard 0.7 adds a fine-tuned model and rewrite blocks
Laya guardLaya Guard 0.7.0 and 0.7.1 (Nov 2-3, 2026) run a fine-tuned Laya model, block rewritten rm -rf forms and route credential reads to the rules.
Category: Coding agents
Pricing: Open source
