<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/where-should-an-agent-s-context-live-a-conversation-with-moonshot-ai-wclpvnkdt" -->

---
title: Where Should an Agent's Context Live? A Conversation...
description: Agent context should split between the model's window and external systems, according to a fireside chat between Moonshot AI's Feihu Tang and PhoenixAI's Sida...
canonical: https://daily.dev/posts/where-should-an-agent-s-context-live-a-conversation-with-moonshot-ai-wclpvnkdt
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Where Should an Agent's Context Live? A Conversation with Moonshot AI | daily.dev
og:description: Agent context should split between the model's window and external systems, according to a fireside chat between Moonshot AI's Feihu Tang and PhoenixAI's Sida...
og:url: https://daily.dev/posts/where-should-an-agent-s-context-live-a-conversation-with-moonshot-ai-wclpvnkdt
og:image: https://api.daily.dev/og/posts/WcLPvnkDT.png
og:image:alt: Where Should an Agent's Context Live? A Conversation with Moonshot AI
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Where Should an Agent's Context Live? A Conversation with Moonshot AI

**[PhoenixAI](https://daily.dev/sources/phoenixai)** · 8 min read · 0 upvotes · 0 comments

## Summary

Agent context should split between the model's window and external systems, according to a fireside chat between Moonshot AI's Feihu Tang and PhoenixAI's Sida Shen. Even with million-token windows, Tang says harder tasks will need more, so retrieval and external context stay necessary. Shen argues database statistics, business definitions and agent traces belong outside the model in a persistent, governed store. Kimi's Agent Swarm shards context so sub-agents return only key conclusions to the orchestrator, and Kimi Code can export session context to Markdown. Cursor trained its Composer model on a Kimi model.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://www.phoenixdata.ai/blog/where-should-agent-context-live-moonshot-ai>

## Questions this post answers

### Does having a million-token context window remove the need for retrieval or external context in AI agents?

No. Moonshot AI's Feihu Tang says that no matter how large a model's context window gets, harder tasks will always need more, so the model's own window and an external harness will keep working together. A million-token window does not eliminate the need for retrieval or an external context layer.

_Anyone weighing retrieval versus bigger context windows for an agent build follows these architecture debates on daily.dev._

### How does Kimi's Agent Swarm handle context between sub-agents and the orchestrator?

Each sub-agent keeps its own working memory and only sends key conclusions back to the master agent, which does not save the full trace. Handling varies by harness: some compress context, some don't, and some compress at set points. Kimi Code includes an export command that writes session context to a Markdown file, showing what was kept.

_Developers designing multi-agent context strategies can track approaches like this on daily.dev._

### What model did Cursor use to train its Composer model?

Cursor trained its Composer model on a Kimi model and then fine-tuned it for its own product, according to Moonshot AI's Feihu Tang. Moonshot releases open weights partly so that other companies can build on them this way, alongside giving some enterprise partners day-zero access to new models.

_Developers comparing coding-agent models can keep up with who trains on whom via daily.dev._

## Similar posts on daily.dev

- [Context Windows Are Not Memory: What AI Agent Developers Need to Understand](https://daily.dev/posts/context-windows-are-not-memory-what-ai-agent-developers-need-to-understand-vkkkaazrp) · Machine Learning Mastery · 3 upvotes · 0 comments
- [AI Agent Context: What Goes Into the Window](https://daily.dev/posts/ai-agent-context-what-goes-into-the-window-2tmrdwqtp) · Redis · 0 upvotes · 0 comments
- [Systemic Approach to Memory, Knowledge, and Context in Agentic AI Architectures](https://daily.dev/posts/systemic-approach-to-memory-knowledge-and-context-in-agentic-ai-architectures-aiwmhrbzc) · InfoQ · 1 upvotes · 0 comments
- [The Context Tax of Agentic Development](https://daily.dev/posts/the-context-tax-of-agentic-development-spfv0nbhr) · Expedia Group Tech · 0 upvotes · 0 comments

---

Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#llm](https://daily.dev/tags/llm), [#rag](https://daily.dev/tags/rag), [#kimi](https://daily.dev/tags/kimi), [#moonshot-ai](https://daily.dev/tags/moonshot-ai)

[View this post on daily.dev](https://daily.dev/posts/where-should-an-agent-s-context-live-a-conversation-with-moonshot-ai-wclpvnkdt)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"Where Should an Agent's Context Live? A Conversation with Moonshot AI","url":"https://daily.dev/posts/where-should-an-agent-s-context-live-a-conversation-with-moonshot-ai-wclpvnkdt","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/where-should-an-agent-s-context-live-a-conversation-with-moonshot-ai-wclpvnkdt"},"datePublished":"2026-11-22T06:35:06.702Z","dateModified":"2026-11-22T06:35:34.860Z","description":"Agent context should split between the model's window and external systems, according to a fireside chat between Moonshot AI's Feihu Tang and PhoenixAI's Sida...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/25face888b14b5a1d56ea2bd1be1c086?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/25face888b14b5a1d56ea2bd1be1c086?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"PhoenixAI","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"PhoenixAI","logo":"https://media.daily.dev/image/upload/s--Bq1aBkxU--/f_auto,pg_1/v1792548289/squads/3e63b4ca-c949-4689-9853-090017390747?_a=BAMAMicg0","url":"https://daily.dev/squads/phoenixai"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/where-should-an-agent-s-context-live-a-conversation-with-moonshot-ai-wclpvnkdt","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,llm,rag,kimi,moonshot-ai","timeRequired":"PT8M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"PhoenixAI","item":"https://daily.dev/squads/phoenixai"},{"@type":"ListItem","position":3,"name":"Where Should an Agent's Context Live? A Conversation with Moonshot AI"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/where-should-an-agent-s-context-live-a-conversation-with-moonshot-ai-wclpvnkdt#faq","mainEntity":[{"@type":"Question","name":"Does having a million-token context window remove the need for retrieval or external context in AI agents?","acceptedAnswer":{"@type":"Answer","text":"No. Moonshot AI's Feihu Tang says that no matter how large a model's context window gets, harder tasks will always need more, so the model's own window and an external harness will keep working together. A million-token window does not eliminate the need for retrieval or an external context layer. Anyone weighing retrieval versus bigger context windows for an agent build follows these architecture debates on daily.dev."}},{"@type":"Question","name":"How does Kimi's Agent Swarm handle context between sub-agents and the orchestrator?","acceptedAnswer":{"@type":"Answer","text":"Each sub-agent keeps its own working memory and only sends key conclusions back to the master agent, which does not save the full trace. Handling varies by harness: some compress context, some don't, and some compress at set points. Kimi Code includes an export command that writes session context to a Markdown file, showing what was kept. Developers designing multi-agent context strategies can track approaches like this on daily.dev."}},{"@type":"Question","name":"What model did Cursor use to train its Composer model?","acceptedAnswer":{"@type":"Answer","text":"Cursor trained its Composer model on a Kimi model and then fine-tuned it for its own product, according to Moonshot AI's Feihu Tang. Moonshot releases open weights partly so that other companies can build on them this way, alongside giving some enterprise partners day-zero access to new models. Developers comparing coding-agent models can keep up with who trains on whom via daily.dev."}}]}
```

