A memory that grows with your agents
mind gives coding agents durable memory across sessions without sending project context to a hosted memory service. It remembers the decisions that matter, recalls them when they become relevant, consolidates repeated knowledge, and gives you explicit control over what stays, changes, or disappears.
Version 7 is the first stable release of the expanded memory platform.
What is new
- Automatic memory: agents can capture durable project context as they work, while transient task state, credentials, and personal identity patterns are rejected or quarantined.
- Better organization: memories can be typed, scoped, pinned, expired, connected to entities, and checked for slot-level contradictions.
- Cross-session continuity: project memory, explicit user memory, provenance, and consolidation receipts make recall explainable instead of opaque.
- Complete control: forget, unlink, redact, and purge cover the full privacy lifecycle, including managed backups and derived files.
- Resilient storage: backup, checkpoint, restore, compaction, deterministic journal merge, and crash-resumable maintenance protect long-lived memory.
- Agent-neutral integration: one local file works with coding agents directly and also exposes structured integration recipes and a standard input/output protocol server.
- Portable by design: the runtime remains offline, deterministic, dependency-free, and distributed as one auditable Python file.
Verified release
- 381 tests passed locally, with one Windows-only case covered by the platform matrix.
- 9 of 9 integration cells passed on Ubuntu, macOS, and Windows with Python 3.9, 3.12, and 3.14.
- Release dry run passed from the signed tag.
- Multilingual recall: 24 of 24.
- Competing-fact discrimination: 12 of 12.
- Slot conflict classification: 50 of 50, with no false positives or false negatives.
- Quick fuzzing: 120 graph cases and 40 command-line cases, with zero failures.
- Mutation evidence: 43 of 120 killed for the single-file runtime and 35 of 120 for the LongMemEval harness, with zero timeouts or infrastructure errors.
- A separate clean-room installation verified the signed tag, byte identity, checksum, initialization, recall, diagnostics, forgetting, redaction, and complete purge.
Install
curl -fsSLO https://raw.githubusercontent.com/Da7-Tech/mind/v7.0.0/mind.py
python3 -c "import hashlib; p=open('mind.py','rb').read(); assert hashlib.sha256(p).hexdigest() == 'ae2fc389b3b09c93cb432ab55b71063d98b400da6b18d6bc178322bc8f3fcf69'"
python3 mind.py initIntegrity
- Tag:
v7.0.0 - Commit:
ade5b2a20e37773e2ae58640b13b59840b739177 mind.pySHA-256:ae2fc389b3b09c93cb432ab55b71063d98b400da6b18d6bc178322bc8f3fcf69
The attached mind.py is byte-identical to the file at the signed tag. SHA256SUMS is included for independent verification. See CHANGELOG.md and the committed raw benchmark reports for the complete engineering record.