[Bug]: Sample configs, install guides and the Helm chart ship message_capacity: 500, which is characters, so every turn triggers a short-term summary #1772

Describe the bug

short_term_memory.message_capacity is a length in characters. The product default is 64,000, and docs/open_source/configuration.mdx documents it that way. But the shipped sample configurations, install guides and Helm chart set it to 500, with no unit.

At 500, a single question and answer overflows the short-term window, so every store starts an LLM summary, and every search waits for a summary in progress before it reads. On an otherwise idle host, searches took 7–32 s and stores about 7 s at the median.

Where 500 ships on main:

• sample_configs/configuration.event.yml:43
• sample_configs/episodic_memory_config.cpu.sample, .gpu.sample, .nebula.sample
• deployments/helm/templates/memmachine-configmaps.yaml:45 and deployments/helm/README.md:135
• docs/install_guide/install_guide.mdx:115 and :232
• docs/install_guide/cloud_deploy/aws_cloudformation.mdx:485
• evaluation/retrieval_agent/README.md (5 places)

Two descriptions also call the value a message count, which makes 500 look reasonable:

• packages/server/src/memmachine_server/episodic_memory/short_term_memory/short_term_memory.py:59: "The maximum number of messages to summarize."
• packages/client/src/memmachine_client/config.py:393: "Maximum number of messages to keep in short-term memory"

Steps to reproduce

1. Deploy with any of the files above (short-term memory enabled).
2. Store a few chat-sized turns (a question and an answer, about 500 characters together) in one session, searching between them.
3. Watch the server's language_model_* metrics: every store triggers a summary call, and searches in that session take seconds.
4. Set message_capacity: 64000 and repeat: no summary calls while the conversation fits, and searches return in well under a second.

Expected behavior

Shipped configurations use a value that holds a conversation (the documented 64,000, or omit the key), and every description of the setting says it is measured in characters.

Environment

• OS: Linux (Ubuntu 24.04), Docker
• MemMachine Version: main at c99bc0e (0.3.9+50.gc99bc0e), image built from source; also seen at c08cf26
• Development language version: Python 3.12 (in the image)
• Backend: event memory, PostgreSQL 18, Qdrant 1.19.1

Additional context

Suggested fix: change the value (or remove the key) in every file listed, correct the two descriptions, and consider renaming the setting or rejecting values smaller than one typical message.