RememberStackremember.dev/docs

Getting started

Give your AI coding agent (Cursor, Claude Code, Claude Desktop, Codex, Antigravity) persistent, auditable memory in under two minutes.

Remember separates raw source claims (testimony) from current truth (adjudicated facts), resolves contradictions autonomously, and gives your agents tools to verify what changed and why.

1. Universal Setup (30 seconds)

Run the interactive setup command in your terminal or project root:

uvx remember setup

(Alternatively, if you prefer pip: pip install remember && remember setup)

What remember setup does:

  1. Connects your backend:
    • Remember Cloud (recommended for instant setup): Connect to https://api.remember.dev with managed private database pods.
    • Self-Hosted Engine: Connect to your local Docker Compose instance (ghcr.io/writeitai/remember-stack:0.17.0) at http://localhost:8000.
  2. Auto-configures your AI agent harnesses:
    • Cursor: Creates .cursor/mcp.json and a .cursor/rules/remember.mdc context rule.
    • Claude Code: Registers the memory server via claude mcp add.
    • Claude Desktop: Updates claude_desktop_config.json.
    • Codex: Updates .codex/config.toml.
    • Antigravity: Generates .agents/mcp_config.json and .agents/skills/remember/SKILL.md.
  3. Guarantees reliable execution: Resolves absolute launcher binaries so your MCP server boots reliably across GUI restarts, without PATH inheritance issues or credential leaks in git.

2. Verify with remember doctor

Verify your setup, backend connectivity, active project, and MCP harness registrations:

remember doctor

Example healthy output:

Remember Diagnostics (CLI v0.17.0)
----------------------------------------
[✓] Configuration: ~/.config/remember/credentials.json (0600)
[✓] Backend connectivity: https://api.remember.dev (200 OK)
[✓] Authentication: Valid session (project: prj_prod_acme)
[✓] MCP Server: uvx remember mcp runnable
[✓] Harnesses configured: Cursor (.cursor/mcp.json), Claude Code
All systems operational.

3. Ingest Your First Document

Add knowledge to your project memory. You can ingest via the CLI or Python SDK:

Via Platform CLI

# Ingest any markdown or text document
printf "# Team Roster\nAlice is the CFO of Acme as of 2026.\n" > roster.md
remember ingest roster.md

Via Python SDK

pip install remember
from remember import RememberClient
 
# Connects using environment variables (REMEMBER_DATA_PLANE_URL, REMEMBER_API_KEY)
# or explicit parameters. Defaults to local self-hosted engine (http://localhost:8000).
client = RememberClient()
 
result = client.ingest(
    filename="roster.md",
    content=b"# Team Roster\nAlice is the CFO of Acme as of 2026.\n",
    mime="text/markdown",
)
print(f"Ingested doc_id: {result.doc_id}, version_id: {result.version_id}")
 
# Wait for extraction, entity resolution, and graph indexing (usually 1-3 seconds)
client.wait_for_readiness([result.version_id], timeout=30)
print("Document is ready for recall!")

4. Query Memory

A. In Your AI Agent

Now open your configured coding agent (Cursor, Claude, Codex, or Antigravity) and ask naturally:

"Check Remember memory: who is the CFO of Acme?"

Your agent will call the facts_context MCP tool, inspect the adjudicated facts and source testimony, and return an honest answer backed by verified evidence.

B. Via Platform CLI

# Get current fact context with live evidence
remember operations run facts_context --arg query="Who is the CFO of Acme?"
 
# Run open SQL against the memory_v1 view
remember query sql "SELECT fact_id, predicate, fact_label FROM facts_current LIMIT 10;"

C. Via Python SDK

# Query current facts
envelope = client.facts_context("Who is the CFO of Acme?")
for fact in envelope.facts:
    print(f"- {fact.label} (evidence count: {fact.evidence_count})")
 
# Run open bitemporal SQL
query_res = client.open_query("SELECT fact_id, predicate, fact_label FROM facts_current LIMIT 10")
print(f"Returned {len(query_res.rows)} facts")

5. Next Steps