Import your conversation history

Five minutes to a useful agent. Export your old AI conversations, run one command, and your next agent starts with everything the last one already knew.

ai-memory mine parses Claude, ChatGPT, and Slack exports into ranked, tiered, recall-ready memories. No re-typing. No copy-paste. No loss of context across model changes.

Why

Why retroactive import matters.

Every AI conversation you've had is a body of decisions, corrections, preferences, and small facts your future agents will need. Today, when you switch tools or start a new context window, that knowledge evaporates. mine ingests it into your local SQLite store — and from that moment, every MCP-compatible AI you talk to recalls it.

This is the fastest path from "new tool, blank slate" to "new tool, every project context I've ever explained."

Onboarding

Five-minute onboarding.

# 1. Drop your export into the working directory
ls conversations.json   # ChatGPT export
# or
ls claude-export/conversations.jsonl   # Claude export
# or
ls slack-export.zip   # extract first → ls slack-export/

# 2. Dry-run to see what will land
ai-memory mine ./conversations.json --format chatgpt --dry-run

# 3. Import for real
ai-memory mine ./conversations.json --format chatgpt

# 4. Confirm + recall
ai-memory stats
ai-memory recall "what database did we decide on for the analytics service"

That's it. Your next agent starts with an opinion.

What gets imported

Each conversation becomes one memory.

Each conversation that survives the --min-messages threshold (default 3) becomes one memory:

Field	Source
title	First user message, truncated
content	Full conversation text — alternating user/assistant turns
tier	mid by default (7-day TTL, auto-promotes to long after 5 recalls)
namespace	Auto-set per format: claude-export, chatgpt-export, slack-export
source	mine-claude / mine-chatgpt / mine-slack
metadata.mined_from	Source format tag
metadata.agent_id	The caller's identity (you)
created_at	Original conversation timestamp

Every imported memory is fully integrated with the rest of the system: scope visibility, governance gates, recall scoring, deduplication on (title, namespace), link creation, consolidation. mine is not a sidecar — it's a first-class write path.

Per-format recipes

Claude export.

Claude exports give you a conversations.jsonl file (one JSON object per line — one conversation per line).

Where to find it:

1. Open Claude → Settings → Privacy → Export data
2. Wait for the email; download the zip
3. Inside, find conversations.jsonl

Import:

# Dry run first — see what will be imported, count + sizes
ai-memory mine ./claude-export/conversations.jsonl --format claude --dry-run

# Real import, default mid-tier
ai-memory mine ./claude-export/conversations.jsonl --format claude

# Or pin everything to long-tier (no TTL) into a custom namespace
ai-memory mine ./claude-export/conversations.jsonl \
  --format claude \
  --namespace personal/claude-history \
  --tier long \
  --min-messages 5

ChatGPT export.

ChatGPT exports give you a conversations.json file (a single JSON array of conversations).

Where to find it:

1. Open ChatGPT → Settings → Data Controls → Export data
2. Wait for the email; download the zip
3. Inside, find conversations.json

Import:

# Dry run
ai-memory mine ./chatgpt-export/conversations.json --format chatgpt --dry-run

# Real import
ai-memory mine ./chatgpt-export/conversations.json --format chatgpt

# Filter to substantive conversations only (10+ messages)
ai-memory mine ./chatgpt-export/conversations.json \
  --format chatgpt \
  --min-messages 10 \
  --tier long

Slack export.

Slack workspace exports give you a directory of per-channel .json files. mine walks the tree.

Where to find it:

1. Slack → Workspace settings → Import / Export Data → Export
2. Pick "Standard export"; wait for the email
3. Download + unzip the archive

Import:

# Dry run on the whole export
ai-memory mine ./slack-export/ --format slack --dry-run

# Real import — narrow to a high-signal channel
ai-memory mine ./slack-export/eng-decisions/ --format slack \
  --namespace work/slack/eng-decisions \
  --tier long \
  --min-messages 4

Flags

Flags reference.

Flag	Default	Meaning
--format, -f	(required)	claude, chatgpt, slack
--namespace, -n	format-specific	Override target namespace
--tier, -t	mid	short / mid / long
--min-messages	3	Skip conversations shorter than this
--dry-run	false	Print what would be imported, don't write

The miner stamps metadata.agent_id to the caller's identity (see Agent identity) and records the original conversation timestamp on created_at.

After

After the import.

Your imported memories are immediately part of the live recall mix:

# Recall ranks across imported + native memories together
ai-memory recall "kubernetes deployment strategy" --tier long

# Filter to just the imported set
ai-memory list --namespace chatgpt-export
ai-memory list --source mine-chatgpt

# Promote the standout decisions to long-tier permanently
ai-memory promote <id> --tier long --priority 9

# Consolidate two redundant decisions into one canonical memory
ai-memory consolidate <id1> <id2> -T "Auth strategy — final decision"

Pattern

A pattern that works.

1. Mine your last 90 days first. Set --min-messages 5 to skip the trivial chatter.
2. Pin the survivors. Run ai-memory recall for your most common project topics; promote the top hits to long-tier.
3. Consolidate the duplicates. Three memories about the same auth flow? ai-memory consolidate collapses them — and keeps the lineage in metadata.consolidated_from_agents.
4. Wire your next AI. Whatever comes next — Claude, GPT, Grok, a local Llama — points at the same SQLite file and inherits everything you imported.

This is the workflow Mem0 charges you SaaS pricing for. ai-memory does it locally, in one binary, in five minutes.

Next

Where to go.

• → Quickstart — store + recall basics
• → Memory tiers — when to use short / mid / long
• → Recall — full scoring + ranking (in the User Guide)
• → Agent identity — every memory carries provenance