Forge Memory Guide

This guide explains how project-scoped memory works in the current fluid forge flow.

Current public entry point

Use fluid forge, not the older fluid forge --mode copilot syntax from archived examples.

Examples:

fluid forge --llm-provider openai --llm-model gpt-4o-mini
fluid forge --no-memory
fluid forge --show-memory
fluid forge --reset-memory

Memory file location

Project-scoped memory lives at:

runtime/.state/copilot-memory.json

It is:

  • repo-local
  • bounded
  • advisory
  • separate from validation

What memory does

When present, memory helps Forge remember stable project conventions such as:

  • preferred provider
  • preferred template or scaffold shape
  • domain hints
  • prior build engines
  • binding conventions
  • bounded schema summaries

Precedence

Forge treats memory as a soft preference.

The effective precedence is:

  1. explicit CLI flags and context input
  2. current-run answers
  3. current discovery
  4. saved project memory
  5. built-in defaults

What is not stored

Memory is not a raw session dump. It does not store:

  • API keys
  • tokens
  • raw sample rows
  • full SQL
  • full README text
  • full contract bodies

Saving and bypassing memory

  • Use --no-memory to ignore saved memory for a run
  • Use --save-memory to persist memory in non-interactive automation
  • Use --show-memory to inspect the remembered summary
  • Use --reset-memory to clear the saved file