- Category
- AI Agents
- Rank
- No. 3850Tools index
Previous survey · No. 3649 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Agent Building
- GitHub
- 14 stars
- Date
About
Speck is a cognitive runtime meant to make small language models more capable by moving memory, attention, planning, evidence tracking, confidence and learning out of the prompt and into persistent, deterministic software. Because cognitive state lives in the runtime, a worker can swap models, restart or move to another compute device without losing task state. It is built on the Genesis runtime.
At a glance
Speck adds persistent memory and optional checks around a small local language model. It is worth exploring if you want the same model to remember corrections across conversations and can test whether that improves answers.
Good fit
Developers comparing a local model with and without a layer that recalls facts and reviews replies.
Consider something else when
Skip it if each conversation must start fresh or you cannot retain user details on disk.
Start here
Use synthetic conversations. Teach it one harmless fact, correct it later, and ask about it in a fresh conversation. Compare the answer with the plain model and inspect what the memory layer recalled.
Cost and requirements
MIT-licensed project with npm setup and a compatible model runtime. Running the model locally requires your own compute.
Limits to know
Stored facts can become stale or add irrelevant context. Review what the system retains and compare recall on realistic examples before enabling it for sensitive conversations.
Sources checked 2026-11-05. Source-based assessment.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.