Prove what your agent did.

    Daryl Sharding MemoryDaryl's flagship product. Off-chain trust, memory and audit layer. Append-only, hash-chained, tamper-evident.

    Shard

    Append entries to named shards — each an append-only, isolated log.

    Index

    Fast, deterministic retrieval through an indexed read layer over your shards.

    Verify

    Ensure consistency with append-only logging and a verifiable SHA-256 hash chain.

    How it works

    1

    Shard the trail

    Append incoming entries to shards — append-only logs you name by session, agent, or domain. Nothing is modified or deleted.

    2

    Index for lookup

    A deterministic read layer indexes entries for fast lookup, kept in sync with shard state. No embeddings, no probabilistic recall.

    3

    Verify State

    Validate data consistency across all shards. Every write is append-only and immutable for auditability.

    4

    Retrieve deterministically

    Query the trail through the read layer. Results return by recency and exact match — the same query over the same data always returns the same result.

    Built for accountable agents

    Multi-Agent Coordination

    Coordinate distributed agents with shared verifiable state. Each agent reads from context, writes to append-only logs.

    Session Continuity

    Persist knowledge across sessions. Agents resume work with complete context — every step still auditable.

    Auditable History

    Every interaction extends the hash-chained trail. History is preserved, append-only, and tamper-evident.

    Design Principles

    Append-only History

    Every state change is immutable and timestamped. Full audit trail with no rewrites.

    Sharded by Context

    Entries organized into distinct, named shards for efficient querying and isolation.

    Verifiable Integrity

    Hash-chained entries and execution traces enable cryptographic proof of integrity (tamper-evidence) — not proof of correctness or truth.

    Deterministic Retrieval

    Same query over same data always returns the same results. No randomness. Grounded answers.

    What DSM is NOT

    Not Chat History

    DSM is a structured, append-only audit trail, not unorganized conversation logs.

    Not a Vector DB

    It is a proof layer, not a retrieval engine. It sits alongside your logs and vector DB — the part you hand to an auditor.

    Not a Black Box

    Every entry is independently verifiable. Full trace of what was stored, when, and chained to what.

    Example Flow

    How DSM records and verifies an agent decision

    1

    Act

    Agent takes an action — the intent is appended to a shard.

    2

    Hash

    SHA-256(content + prev_hash) chains the entry to all prior ones.

    3

    Sign

    Optional Ed25519 signature proves authorship.

    4

    Seal

    Close the shard with a cryptographic tombstone for archival.

    5

    Verify

    Replay the chain, recompute every hash, confirm integrity — independently.