memcite 0.2.0

Repo memory for AI agents — every memory has a source, every source gets verified.

agentic-memory (memcite)

Open-source repo memory for AI agents — every memory has a source, every source gets verified.

Why

AI agents forget everything between sessions. Existing memory layers (mem0, Zep, LangMem) store text in vector DBs but can't tell you where that knowledge came from or whether it's still true.

agentic-memory enforces a simple rule: No evidence, no memory.

Every memory must cite its source (file path + line number, git commit, URL). Before an agent uses a memory, the citation is automatically re-validated. Stale memories get flagged, not silently served.

How it works

from agentic_memory import Memory, FileRef

mem = Memory("./my-project")

# Store a memory — citation is required
mem.add(
    "This project uses ruff for linting with line-length=120",
    evidence=FileRef("pyproject.toml", lines=(15, 20)),
)

# Query — returns answer + citation status
result = mem.query("What linter does this project use?")
print(result.answer)     # "ruff with line-length=120"
print(result.citations)  # [FileRef("pyproject.toml", L15-20, status=VALID)]

# Validate all memories — find what's gone stale
stale = mem.validate()
# [StaleMemory("ruff config", reason="file content changed at L15")]

Design Principles

1. No Evidence, No Memory — add() without a citation raises an error
2. Validate Before Use — query() re-checks citations by default
3. Decay What's Stale — confidence drops when evidence changes; invalid memories are deprioritized

Evidence Types

Type	What it tracks	Validation method
FileRef	File path + line range	Check file exists, content matches
GitCommitRef	Commit SHA + file	Verify commit exists in history
URLRef	Web URL	HTTP HEAD check + content hash
ManualRef	Human-provided note	No auto-validation (always trusted)

Features

• Repo-scoped — each repository gets its own memory namespace
• Local-first — SQLite storage, no external services required
• Citation-backed — every memory traces back to a verifiable source
• Auto-validation — stale evidence is detected before it misleads your agent
• Confidence scoring — memories with invalid citations get deprioritized
• CLI included — am add, am query, am validate, am status

Installation

pip install memcite

With extras:

pip install memcite[mcp]     # MCP server for Claude Code
pip install memcite[api]     # REST API server (FastAPI)

CLI Usage

# Add a memory with file evidence
am add "Uses pytest for testing" --file tests/conftest.py --lines 1-10

# Query memories
am query "What test framework?"

# Validate all memories
am validate

# Show memory status
am status

MCP Server

Add to your .mcp.json to use with Claude Code:

{
  "mcpServers": {
    "agentic-memory": {
      "command": "am-mcp",
      "args": ["--repo", "/path/to/your/project"]
    }
  }
}

Tools: memory_add, memory_query, memory_validate, memory_status, memory_list, memory_delete

REST API

am-server --repo /path/to/repo --port 8080

OpenAPI docs at http://localhost:8080/docs. Endpoints:

Method	Path	Description
POST	/memories	Add a memory with evidence
POST	/memories/query	Hybrid search + citation validation
GET	/memories	List all memories
GET	/memories/{id}	Get a specific memory
DELETE	/memories/{id}	Delete a memory
POST	/memories/validate	Validate all citations
GET	/status	Memory status summary

Hybrid Search

When initialized with an embedding provider, queries combine FTS5 full-text search with vector similarity:

from agentic_memory import Memory, TFIDFEmbedding, FileRef

mem = Memory("./my-project", embedding=TFIDFEmbedding())
mem.add("Uses ruff for code formatting", evidence=FileRef("pyproject.toml", lines=(1, 5)))

# Finds the memory even though "linting" != "formatting"
result = mem.query("What linter does this project use?")

Default weights: FTS5 (0.65) + Vector (0.35). Customize per query:

result = mem.query("linting", fts_weight=0.5, vector_weight=0.5)

Admission Control

Filter out low-value memories before they're stored:

from agentic_memory import Memory, HeuristicAdmissionController

mem = Memory("./my-project", admission=HeuristicAdmissionController())
mem.add("ok", evidence=ManualRef("chat"))  # raises ValueError — too vague

Or use LLM-based scoring with any OpenAI-compatible API:

from agentic_memory import LLMAdmissionController

def my_llm(system: str, user: str) -> str:
    # Call your LLM here, return JSON: {"score": 0.0-1.0, "reason": "..."}
    ...

mem = Memory("./my-project", admission=LLMAdmissionController(llm_callable=my_llm))

Roadmap

• Core SDK — add / query / validate with citation enforcement
• CLI tool
• MCP Server — use with Claude Code and other MCP clients
• Admission control — LLM-based scoring to filter low-value memories
• Hybrid search — FTS5 + TF-IDF vector fusion, pluggable embedding providers
• REST API server — FastAPI with OpenAPI docs
• LangChain / LlamaIndex integration
• Web dashboard

Compared to

	mem0	Zep	LangMem	agentic-memory
Vector search	Yes	Yes	Yes	Yes
Forced citations	No	No	No	Yes
Source validation	No	No	No	Yes
Staleness detection	No	No	No	Yes
Repo-scoped	No	No	No	Yes
Self-hosted	Yes	Yes	Yes	Yes

License

MIT