Memvid
MemvidMemvid packages an agent's memory into a single portable .mv2 file holding content frames, full-text and vector indexes and a time index, so retrieval runs locally without a database server. An enrichment step extracts entity facts called memory cards and records how each value changes over time. Memvid, Inc. builds it as a Rust core with Python, Node.js and command-line packages.
Openness
4 medium confidence- license
- Apache-2.0 (memvid-core, the Rust library and file format)
- source
- public (memvid/memvid publishes the Rust core; the Python and Node.js SDK bindings and the CLI ship as compiled packages, and no public memvid repository holds their source)
- core features withheld
- yes — the core embeds Memvid's ticket public key and caps a new file at the 50 MB free tier; larger capacity needs a ticket signed by Memvid's control plane for a paid Starter, Pro or Enterprise plan
- commercial
- Memvid dashboard and capacity plans
Memvid's Rust core is published under Apache-2.0, but it limits each memory file to a small free capacity and raises that limit only with tickets signed by Memvid's servers for paying plans. The Python and Node.js SDKs and the command-line tool are distributed as compiled packages whose native code has no published source.
- https://api.github.com/search/repositories?q=org:memvid&per_page=50 recorded 2026-11-20
The memvid organization's public repositories are memvid, example apps, a benchmark tool, a Homebrew tap and memory-cli, "A simple Rust CLI used to test Homebrew, Winget, and APT release pipelines"; none holds the SDK bindings or the memvid CLI.
- https://docs.memvid.com/concepts/capacity-and-plans.md recorded 2026-11-20
The plans table lists Free at 50 MB, Starter at 25 GB for $9.99/mo, Pro at 125 GB for $49.99/mo and Enterprise unlimited; to increase capacity, get an API key from the dashboard and sync tickets to the local file.
- https://raw.githubusercontent.com/memvid/memvid/main/Cargo.toml recorded 2026-11-20
The manifest names the crate memvid-core, "Core library for Memvid v2, a crash-safe, deterministic, single-file AI memory", with license = "Apache-2.0" and repository github.com/memvid/memvid.
- https://raw.githubusercontent.com/memvid/memvid/main/LICENSE recorded 2026-11-20
The root LICENSE is the Apache License, Version 2.0 text.
- https://raw.githubusercontent.com/memvid/memvid/main/src/memvid/ticket.rs recorded 2026-11-20
apply_signed_ticket verifies a ticket against the embedded MEMVID_TICKET_PUBKEY before raising capacity: "Only tickets signed by the official Memvid control plane will be accepted."
- https://raw.githubusercontent.com/memvid/memvid/main/src/types/common.rs recorded 2026-11-20
Tier::Free has a capacity of 50 * 1024 * 1024 bytes, commented as 50 MB.
Adoption
2 medium confidenceAdoption is read from monthly downloads of Memvid's four packages, the Node.js SDK and command-line tool on npm, the Python SDK on PyPI and the Rust crate, with the crate's 90-day count converted to a monthly rate. The older memvid package on PyPI is the retired video-based first version and is left out.
- https://api.npmjs.org/downloads/point/last-month/@memvid/sdk recorded 2026-11-20
30,899 downloads of @memvid/sdk between 2026-10-20 and 2026-11-18.
- https://api.npmjs.org/downloads/point/last-month/memvid-cli recorded 2026-11-20
8,865 downloads of memvid-cli between 2026-10-20 and 2026-11-18.
- https://crates.io/api/v1/crates/memvid-core recorded 2026-11-20
crate memvid-core: 26,813 downloads all-time and 17,318 recent_downloads (trailing 90 days); repository github.com/memvid/memvid; latest version 2.0.140.
- https://pypi.org/pypi/memvid/json recorded 2026-11-20
The memvid package on PyPI is version 0.1.3, "Video-based AI memory library for fast semantic search and retrieval", with its homepage at github.com/olow304/memvid.
- https://pypistats.org/api/packages/memvid-sdk/recent recorded 2026-11-20
last_month downloads = 1,868 for memvid-sdk.
Capability
4 medium confidenceMemvid extracts entity facts from stored content, records when a new value replaces an older one, and keeps those facts as typed cards beside an entity graph and the original documents. It does not retire or expire memories on its own: old frames stay in the file until someone deletes them.
- https://docs.memvid.com/cli/index recorded 2026-11-20
The CLI reference groups commands into creation, data, search (find, ask, vec-search, timeline), enrichment (enrich, memories, state, facts), maintenance and sessions; deletion is a delete command the user runs.
- https://docs.memvid.com/concepts/memory-cards.md recorded 2026-11-20
Enrichment extracts entity-slot-value memory cards of kinds fact, preference, event, profile, relationship and goal, using rules, a local Phi-3.5-mini model or hosted LLMs; fact history marks later values "UPDATES previous" and state returns the current value.
- https://docs.memvid.com/introduction/the-memvid-approach recorded 2026-11-20
Each file carries lexical, vector, SimHash, time and Logic Mesh indexes, the last an entity-relationship triple store queried with memvid follow traverse.
- https://raw.githubusercontent.com/memvid/memvid/main/README.md recorded 2026-11-20
The README describes "a single-file memory layer for AI agents with instant retrieval and long-term memory", with BM25 full-text search, HNSW vector search and a time index inside one .mv2 file.
Verified 2026-11-20, except some scores, still awaiting re-confirmation