5 ms·

I'm fed up with compaction. I want my agent to get compacted but also retain full access to the prior conversation via search and tool calls - I want it to know

by simonw 2mo ago

I'm fed up with compaction. I want my agent to get compacted but also retain full access to the prior conversation via search and tool calls - I want it to know "the requirements for X were discussed in detail previously in conversation C51E31CE-C985-4633-A749-DCC9805A7FEB" and have a tool that lets it dispatch a subagent to find those details again. Do any of the coding agents have this already?

weitendorf 2mo ago
Claude Code makes agents reasonably aware of where their log files/history/etc are and get stored. Generally they’ll work with them without explicitly being told (especially to recover broken sub agents, corrupted sessions, etc) to do so. I think the more general problem is that compaction is just a bandaid: you HAVE to dump context to keep going and searching back for it is more expensive than if you had just kept the right context. The better a job the harness does at filtering out junk, the more likely compaction is to remove context that might have been, forgive me, “load bearing” IMO the default Claude Code / Codex (which to my understanding is almost continually-compacting?) compaction has got much better over the part few months. If you spam sub agents then context will naturally nest, and you can just resurrect them as needed without polluting the main thread.
auspiv 2mo ago
“Search session history for when we discussed XYZ” https://github.com/nerdyaustin/memory_mcp https://github.com/nerdyaustin/memory_mcp You can skip all the sync stuff, not necessary at all
wild_egg 2mo ago
This one does agentic search instead of compaction: https://github.com/wedow/harness https://github.com/wedow/harness Generally works quite well but it allows nested subagents and that can use a lot of tokens if the agent prompts are too open-ended. All messages are persisted as separate markdown files under a session directory, which makes them much more friendly to grep and such than the common jsonl files are. Agents can search them rapidly without messing around with jq. Also written in Bash so it runs on any old potato you have laying around.
smagnuso 2mo ago
Hydra is an agent in spirit, but actually wraps other agents through ACP. It drives its own async compaction algorithm that gives the agent it wraps full access to history that it can search as an MCP server after compacting. https://github.com/smagnuso/hydra-acp https://github.com/smagnuso/hydra-acp
porridgeraisin 2mo ago
Both CC as well as grok build for me seem to know about the log file location and they just read off the context from there
prashantsengar 2mo ago
Hydron[1] has a search previous chats tool. It's built for embedded software but works well for general software engineering work too. Disclaimer: I am the lead dev for it. [1] https://www.hydron.sh/ https://www.hydron.sh/
mceachen 2mo ago
Claude is quite good at searching both its own and codex session transcripts, and if asked, can spawn tmux/konsole/whatever to resume those sessions to be interactive if needed. My system prompt adds TPPs and TLDRs for task/feature planning, and shared memories/documentation management. I find them invaluable. https://photostructure.com/coding/claude-code-tpp/ https://photostructure.com/coding/claude-code-tpp/
3371 2mo ago
In my own framework (just a skill), every unit of works produce a markdown document. This provides rich context over past plan/interview/implementation/etc., this + git is more than enough to not think about context and memory.

by simonw 2mo ago

weitendorf 2mo ago

Claude Code makes agents reasonably aware of where their log files/history/etc are and get stored. Generally they’ll work with them without explicitly being told (especially to recover broken sub agents, corrupted sessions, etc) to do so. I think the more general problem is that compaction is just a bandaid: you HAVE to dump context to keep going and searching back for it is more expensive than if you had just kept the right context. The better a job the harness does at filtering out junk, the more likely compaction is to remove context that might have been, forgive me, “load bearing” IMO the default Claude Code / Codex (which to my understanding is almost continually-compacting?) compaction has got much better over the part few months. If you spam sub agents then context will naturally nest, and you can just resurrect them as needed without polluting the main thread.

auspiv 2mo ago

“Search session history for when we discussed XYZ” https://github.com/nerdyaustin/memory_mcp https://github.com/nerdyaustin/memory_mcp You can skip all the sync stuff, not necessary at all

wild_egg 2mo ago

This one does agentic search instead of compaction: https://github.com/wedow/harness https://github.com/wedow/harness Generally works quite well but it allows nested subagents and that can use a lot of tokens if the agent prompts are too open-ended. All messages are persisted as separate markdown files under a session directory, which makes them much more friendly to grep and such than the common jsonl files are. Agents can search them rapidly without messing around with jq. Also written in Bash so it runs on any old potato you have laying around.

smagnuso 2mo ago

Hydra is an agent in spirit, but actually wraps other agents through ACP. It drives its own async compaction algorithm that gives the agent it wraps full access to history that it can search as an MCP server after compacting. https://github.com/smagnuso/hydra-acp https://github.com/smagnuso/hydra-acp

porridgeraisin 2mo ago

Both CC as well as grok build for me seem to know about the log file location and they just read off the context from there

prashantsengar 2mo ago

Hydron[1] has a search previous chats tool. It's built for embedded software but works well for general software engineering work too. Disclaimer: I am the lead dev for it. [1] https://www.hydron.sh/ https://www.hydron.sh/

mceachen 2mo ago

Claude is quite good at searching both its own and codex session transcripts, and if asked, can spawn tmux/konsole/whatever to resume those sessions to be interactive if needed. My system prompt adds TPPs and TLDRs for task/feature planning, and shared memories/documentation management. I find them invaluable. https://photostructure.com/coding/claude-code-tpp/ https://photostructure.com/coding/claude-code-tpp/

3371 2mo ago

In my own framework (just a skill), every unit of works produce a markdown document. This provides rich context over past plan/interview/implementation/etc., this + git is more than enough to not think about context and memory.