What is this?
TierSense is a hub of three research-backed APIs that give an AI agent decision signals about task difficulty, which history steps are ready to compress, and whether to compress or recall memory right now, so your own app can act on those signals.
TierSense is a hub of three research-backed APIs that give an AI agent decision signals about task difficulty, which history steps are ready to compress, and whether to compress or recall memory right now, so your own app can act on those signals.Is this a library I install, or a hosted API I call?It's a hosted API service you call with an API key; the repository is just the documentation and integration hub for the three endpoints, not an SDK or self-hosted server.
It's a hosted API service you call with an API key; the repository is just the documentation and integration hub for the three endpoints, not an SDK or self-hosted server.How do I start using it?Get an API key from the TierSense platform, then send POST requests to the Score, Compress, or Memory-Decide endpoints as needed.
Get an API key from the TierSense platform, then send POST requests to the Score, Compress, or Memory-Decide endpoints as needed.Do I need to call all three APIs?No, each module is independent: use Score alone for difficulty display or model routing, Compress alone to pick history to summarize, or Memory-Decide alone to decide when to recall; combine them only if your use case needs it.
No, each module is independent: use Score alone for difficulty display or model routing, Compress alone to pick history to summarize, or Memory-Decide alone to decide when to recall; combine them only if your use case needs it.What exactly do Score, Compress, and Memory-Decide each output?Score returns five feature scores plus an overall difficulty score, Compress returns per-step compression recommendations and span selection, and Memory-Decide returns two decisions (compress/recall) with confidence values.
Score returns five feature scores plus an overall difficulty score, Compress returns per-step compression recommendations and span selection, and Memory-Decide returns two decisions (compress/recall) with confidence values.What is the story behind the 'Jev' mentions in the README?The authors state their methods and two papers were completed before Jev appeared on September 15, 2026, representing roughly six months of independent work, and they push back on being labeled a rushed 'auto research' response to it.
The authors state their methods and two papers were completed before Jev appeared on September 15, 2026, representing roughly six months of independent work, and they push back on being labeled a rushed 'auto research' response to it.Where can I read the underlying research?Two linked PDFs cover the work: 'Memory Control Signals Emerge Before Action in Long Horizon Agents' (35 pages) and 'StateComp: Learning When to Compress History in Long Horizon Agents' (33 pages).
Two linked PDFs cover the work: 'Memory Control Signals Emerge Before Action in Long Horizon Agents' (35 pages) and 'StateComp: Learning When to Compress History in Long Horizon Agents' (33 pages).Is this the only repository I need, or are there older ones?This repo is now the single maintained home for docs, integration examples, Issues, and Stars; older repositories only keep migration notices and historical commit history.
This repo is now the single maintained home for docs, integration examples, Issues, and Stars; older repositories only keep migration notices and historical commit history.What are similar projects?context-language-models, memcode, mem0, EverOS, Raven, deer-flow, generative_agents, Athena-Public
context-language-models, memcode, mem0, EverOS, Raven, deer-flow, generative_agents, Athena-PublicWhich topics is it in?llm-memory, model-router, arxiv, long-term-memory, llm-agents, papers, agentic, paper, context-engineering, academic, saas, autonomous-agents
llm-memory, model-router, arxiv, long-term-memory, llm-agents, papers, agentic, paper, context-engineering, academic, saas, autonomous-agents