Agent context and memory

Tenant-supplied agent context

Drop markdown files in workspace/.pokee/*.md to prepend tenant-specific instructions to the agent's system prompt. Effective on the next session (in-flight sessions keep the prompt they were created with).

Examples:

• .pokee/soul.md — voice / role / personality
• .pokee/glossary.md — your domain terms
• .pokee/style_guide.md — output formatting conventions

# Upload via the standard files API (use any session id; the file lives in the workspace)
curl -X PUT "$POKEE_API/v1/sessions/$SID/files/.pokee/soul.md" \
  -H "Authorization: Bearer $POKEE_KEY" \
  -H "Content-Type: text/markdown" \
  --data-binary @soul.md

Files load alphabetically. Prefix with 00_, 10_ if order matters. Each file gets a ## <filename> header in the system prompt so the agent can reference it back ("update soul.md so that …"). Total content cap: 1 MB across the whole .pokee/ directory.

Memory across sessions

The agent has a persistent file at workspace/.memory/MEMORY.md (and any sibling files it chooses to create). It's instructed to read this at the start of each session and update it when you share facts worth remembering long-term ("I prefer terse answers", "my project is X"). It survives across sessions and pod restarts.

You can inspect or edit it directly via the files API:

curl "$POKEE_API/v1/sessions/$SID/files/.memory/MEMORY.md" \
  -H "Authorization: Bearer $POKEE_KEY"

If you want strict control over what context the agent carries across sessions (and prefer to manage everything via .pokee/), ask Pokee to disable the memory hint for your tenant.