Folders, docs, and decisions
Indexes your project folders, documents, configuration notes, and architecture decisions so nothing gets buried.
Coming soon
Shared memory for your agents.
Memophant is one searchable knowledge base, served as an MCP server, that every AI coding agent you use can query. An elephant never forgets, so an agent can find what another agent already figured out last week.
Not available yet · €10 one-time at launch
What it remembers
Your files, your decisions, and your past sessions
Memophant brings together the context scattered across your folders and coding tools into one index on your machine.
Full coding transcripts
Indexes the full conversation transcripts of your AI coding sessions across Claude Code, Google Antigravity, and OpenAI Codex.
Synced across computers
Includes sessions synced from your other computers, so an agent today can pick up what another agent solved last week.
Search that answers
Ranked snippets with transparent query relaxation
Full-text search returns ranked snippets with the file path and modification time. Filter by folder, file type, and age; choose whether transcripts are included, excluded, or searched alone. If no passage holds every word, the search relaxes the query step by step and tells you which step answered and the exact query that ran.
query: "sqlite wal checkpoint lock timeout"
step: 2 of 4 (relaxed: dropped "checkpoint")
ran: sqlite AND wal AND lock AND timeout
No passage matched all five terms. Step 2 relaxed the query to
sqlite AND wal AND lock AND timeout and returned ranked hits below.
~/projects/api/notes/storage-decisions.md
modified 6 days ago
Switched connection setup to sqlite wal mode with busy timeout set to 5000ms to clear the write lock during concurrent background jobs.
~/.claude/sessions/2026-11-08-api-fix.jsonl
modified 6 days ago
Verified that enabling wal on sqlite plus a higher busy timeout resolved the database lock errors seen under parallel workers.
How it works
From local files to grounded answers in four steps
-
Index locally
Memophant runs on your own machine and indexes your project folders, notes, and synced transcripts locally while policy checks keep secrets out.
-
Search with relaxation
Your agent queries the MCP server with optional folder, file type, age, and transcript filters. If needed, the query relaxes step by step until it finds matches.
-
Read the source
The agent opens a hit and reads it in pages: only paths returned by a search in that session, and only after eligibility and secret checks pass.
-
Recall without asking
An optional hook adds the most relevant hits to each prompt, so the agent starts every turn with leads from your own history.
Safe by design
Local indexing, secret filtering, and untrusted-data boundaries
Answers come with their source, not a summary to trust blindly, and strict checks sit between your disk and your agents.
Runs on your machine
Memophant runs on your own machine and indexes locally. Your folders and transcripts stay under your control.
Secrets kept out
Files with secrets are excluded by policy checks during indexing and checked again before any hit can be opened.
Session-scoped reads
An agent can only open and page through paths that a search returned in that same session, after eligibility and secret checks pass.
Marked as untrusted data
Everything returned from the knowledge base is marked as untrusted data, so text inside an old document or transcript is never followed as an instruction.
One memory, every agent
Any MCP client draws on the same knowledge base
Because Memophant is served as an MCP server, all of your coding agents share one memory instead of starting from scratch. Index health and size are one call away.
Pricing
Simple one-time license at launch
Not available yet
€10 one-time
One-time payment when Memophant launches. No launch date is set yet.
- Local MCP server and full-text index
- Project folders, notes, and coding session transcripts
- Optional prompt-recall hook and index health check
Tell me at launch
Opens your email app to hello@memophant.dev
FAQ
Questions & answers
What is Memophant?
Memophant is shared memory for your AI coding agents: one local, searchable knowledge base served as an MCP server that every agent you use can query.
Which agents can use it?
Any MCP client can connect to Memophant, so Claude Code, Google Antigravity, OpenAI Codex, and Gemini-based agents all draw on the same memory.
What does Memophant index?
It indexes your project folders, documents, configuration notes, and decisions, plus the full conversation transcripts of your AI coding sessions (Claude Code, Google Antigravity, OpenAI Codex), including sessions synced from your other computers.
What happens if a search query is too specific?
If no passage holds every word, Memophant relaxes the query step by step and tells you which step answered and the exact query that ran.
Can an agent read arbitrary files on my disk?
No. An agent can open a hit and read it in pages only for paths that a search returned in that session, and only after eligibility and secret checks pass.
How does it protect against secrets and prompt injection?
Memophant runs on your own machine and indexes locally. Policy checks keep files with secrets out, and everything returned from the knowledge base is marked as untrusted data so text inside an old document or transcript is never followed as an instruction.
How much does it cost and when does it launch?
Memophant is not available yet, and no launch date has been set. At launch, it will cost €10 one-time. Click “Tell me at launch” to email hello@memophant.dev.