Membase

membase-ai

One memory. Every AI.
Drop-in memory infrastructure for AI agents and apps — the Membase SDK.

Sources go into one memory that every AI and app reads

The Membase SDK: a Python client, the membase command line, an MCP server and a TypeScript client. The memory lives in your Membase account or on your machine, behind the same API.

Write once, read from any AI

Your app writes a memory; any AI reads it, here Claude Code

Recordings: Python SDK (57 s) · Claude Code (1 min 33 s). The answer above is illustrative.

Accuracy

LoCoMo 93.1%, LongMemEval_S 92.6%, DMR 92.2%

On three public long-term memory benchmarks, with a few thousand tokens of context per question.

Quickstart

Hosted, local in process, or local over HTTP

pip install membase-ai              # hosted · Python 3.10+ · needs MEMBASE_API_KEY
pip install 'membase-ai[local]'     # local  · Python 3.12 or 3.13 · needs OPENAI_API_KEY
npm install membase-ai              # TypeScript · Node 18+

from membase import Membase

m = Membase()                    # hosted
# m = Membase(local=True)        # local, in ~/.membase

m.memories.add("We picked Postgres for the ledger service.", container="Engineering")
m.search("what database is the ledger on?")
m.add("Design notes …", container="Engineering", custom_id="design-1")   # a document
m.ask("Which database did we choose for the ledger?")

The package is membase-ai; membase on PyPI is unrelated.

Command line

membase --local add "We picked Postgres for the ledger" --container Engineering
membase --local search "what did we pick for the ledger?"
membase --local ask "Which database is the ledger on?"
membase --local import ~/Downloads/claude-export.json     # Claude, ChatGPT, markdown or JSON chats
membase --local agent ingest trace.json --agent coder
membase --local agent search "fix flaky deploy" --agent coder

Without --local (or --store DIR, both before the command) commands use the hosted API; agent, import and serve need membase-ai[local].

MCP

MCP clients, hosted or local server, same tools

claude mcp add membase -- membase --local mcp                                   # local
claude mcp add --transport http membase https://api.app.membase.io/mcp-http \
  --header "Authorization: Bearer $MEMBASE_API_KEY"                             # hosted

Hosted without the header, the connection signs you in and is read-only.

Reference

Methods

Client methods: read, write, remove, answer

Deleting a document or forgetting a memory needs confirm=True; without it you get what would be removed.
Hosted, ask needs an agent-endpoint key, and each key is limited to its containers and access level.

What the local engine stores

Conversations, documents and agent traces

Local memory over HTTP

membase --local serve            # http://127.0.0.1:8787/v1

The hosted API's agent-protocol routes, answered by the local engine: point any client's base URL at it. Set MEMBASE_LOCAL_TOKEN to require it as a bearer token.

Local settings

Stores: ~/.membase/memory.db (default container), ~/.membase/containers/<id>/ (others), ~/.membase/profile/ (static facts). Settings come from the environment, or from a .env file in the current directory or a parent.

Setting	Default
`OPENAI_API_KEY`		chat and embeddings
`MEMBASE_LLM_PROVIDER`	from your keys	`openai`, `anthropic`, `ollama` or `openai-compat` for chat; set the models below to match
`MEMBASE_LLM_ENDPOINT`		base URL for `ollama` or `openai-compat`
`MEMBASE_EMBED_PROVIDER`	`openai`	`openai-compat` uses `MEMBASE_LLM_ENDPOINT`
`MEMBASE_EMBED_MODEL`	`text-embedding-3-small`	embeddings
`MEMBASE_EPISODE_MODEL`	`gpt-4.1-mini`	conversations into episodes
`MEMBASE_DECIDER_MODEL`	`gpt-4o-mini`	picks the episodes a search returns
`MEMBASE_READER_MODEL`	`gpt-4o`	`ask` answers
`MEMBASE_KNOWLEDGE_MODEL`	`gpt-4o-mini`	documents into topic trees
`MEMBASE_AGENT_MODEL`	`gpt-4o-mini`	agent traces
`MEMBASE_LANG`	`en`	`zh` for Chinese extraction prompts

MEMBASE_LOCAL=1 makes local the default when MEMBASE_API_KEY is not set.

Repository

membase/ is the Python package (client.py, cli.py, local/ for local memory and membase serve, mcp/ for membase mcp); typescript/ is the npm package. membase-sdk (PyPI and npm) is the earlier name, kept as an alias.

License

MIT. membase-ai[local] also installs membase-core, the compiled engine, under Unibase's proprietary license.

Metadata

Release files for membase-ai 0.2.19

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for membase-ai 0.2.19
File	Size	Uploaded
membase_ai-0.2.19.tar.gz	25.6 kB	Oct 6, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for membase-ai 0.2.19
File	Interpreter	ABI	Platform	Reset
membase_ai-0.2.19-py3-none-any.whl 30.7 kB Oct 6, 2026	Python 3	none	any	Details

Total release size: 56.3 kB

Release files / membase_ai-0.2.19.tar.gz

Download URL	membase_ai-0.2.19.tar.gz
Size	25.6 kB
Tags	Source
SHA-256 checksum How to use checksums	`ccbafa4dfb31ebafc53d3db542670e24e63fefc305cd85e79cdbf1d4d2dae8a6`
BLAKE2b-256 checksum How to use checksums	`de7463f42d4d2b2a25404da11cf64fca23da124225b9e8433615d74245367851`
Upload date	Oct 6, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.13.12`

Release files / membase_ai-0.2.19-py3-none-any.whl

Download URL	membase_ai-0.2.19-py3-none-any.whl
Size	30.7 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`4a2b5e6b1efb0e73370e24dd6d04615ab19c5a8c1c947c24075b9e2d37b81ae0`
BLAKE2b-256 checksum How to use checksums	`a2008d80e405cd6541d4e8d1bd345a63c55b37280af92e77915b46718280c642`
Upload date	Oct 6, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.13.12`