agentmem

Production-ready agent memory in one line. No vector DB. No infrastructure. Just SQLite.

[](https://pypi.org/project/agentmem/) [](https://pypi.org/project/agentmem/) [](https://opensource.org/licenses/MIT) [](https://github.com/thezenmonster/agentmem/actions)

pip install agentmem

from agentmem import Memory

mem = Memory()

# Store memories by type
mem.add(type="bug", title="loudnorm undoes SFX reductions",
        content="Never apply loudnorm to final amix output. It lifts noise floor.")

mem.add(type="setting", title="Voice speed config",
        content="atempo 0.90 per-line then 1.15x global.",
        tags=["voice", "audio"])

# Full-text search (instant, FTS5-powered)
results = mem.search("audio mixing")

# Context-budgeted recall (the killer feature)
context = mem.recall("building a narration track", max_tokens=2000)
# Returns: formatted markdown of the most relevant memories,
#          fitted to your token budget, ranked by relevance + recency

Every agent builder hits the same wall: your agent forgets everything between sessions.

Current solutions are either too heavy (vector databases, API keys, cloud services) or too simple (flat files that don't scale). agentmem sits in the sweet spot:

| Feature | Mem0 | Engram | MCP Memory | agentmem | |---|---|---|---|---| | pip install, done | Needs API key or vector DB | Go binary | npm package | Yes | | Typed memories | Blob store | Key-value | Entities/relations | 6 types | | Python API | Yes | Go only | Node.js | Yes | | Full-text search | Vector-based | FTS5 | No search | FTS5 | | MCP server | Separate setup | Built-in | Built-in | Built-in | |