Item detail
github.com

Sibyl-Labs/Sibyl-Memory

RepoRadar surfaced Sibyl-Labs/Sibyl-Memory — an AI project — into the MIT local-first agent-memory family section, where it sits at Silver tier with a 'try now' verdict. Its strongest signal is workflow potential, scored 9.2 out of 10.

Score7.7
Popularity88.0
Riskconditional
TierSilver
Score breakdown
Usefulness8.0
Novelty9.0
Momentum7.0
Maturity7.4
Open-source/build8.4
Evidence8.0
Workflow potential9.2
Setup ease8.8

Popularity is tracked separately. Support, ads, sponsorships, and tips never affect these signals.

Why it matters

Useful for AI agent authors who want a **local-first** memory substrate that runs on a single SQLite file with FTS5 search — no vector store, no embeddings, no retrieval model, no external infrastructure — and the agent runs on a 4 vCPU / 16 GB box; for engineering teams that need **multi-tenant by design** — the schema enforces `UNIQUE (tenant_id, category, name)` at the SQLite level so two

Who should use it

AI agent authors who want a **local-first** memory substrate that runs on a single SQLite file with FTS5 search — no vector store, no embeddings, no retrieval model, no external infrastructure — and the agent runs on a 4 vCPU / 16 GB box Engineering teams that need **multi-tenant by design** — the schema enforces `UNIQUE (tenant_id, category, name)` at the SQLite level so two agents cannot stomp each other's WARM entities Engineering teams that need a **rule-enforced single source of truth per entity** — Rule 43 is enforced at the schema level, not just by convention in the application code, and drift is impossible by construction Engineering teams that want **append-only audit trail** — the COLD journal tier is append-only by schema, ARCHIVE keeps retired entities for audit, REFERENCE is for static knowledge that rarely changes, HOT is the live working state rewritten in place **Hermes Agent v0.13+** users — `pip install sibyl-memory-hermes && sibyl-memory-hermes install-plugin` and edit `~/.hermes/config.yaml` with `memory: {provider: sibyl}` MCP-compatible agent users (Claude Code, Codex, Cursor, Continue) — `pip install sibyl-memory-mcp` and point any MCP client at the bundled server entry point Users who care about **privacy disclosure** — the README is explicit that memory content never leaves the machine, the only outbound network call is tier verification (account id, session token, DB byte size, proposed delta — never the memory contents), and free unactivated use makes no network calls at all Engineering teams evaluating research-vs-production — the 95.6% LongMemEval Oracle score is self-reported and the blog explicitly notes 'no official leaderboard exists, judges and generator models vary across entries,' but the architecture is the only file-based system in the top tier of the community leaderboard Engineering teams that want a **linter** to surface common architectural smells (unbounded HOT writes, duplicate WARM entities) — the memory linter is part of the activated path **Vendor-API-key-prerequisite caveat (cycle 164 pattern)**: the **free, unactivated path** makes no network calls and runs entirely local — this is the right starting surface for evaluation; the **activated path** binds a wallet or email and writes credentials to `~/.sibyl-memory/credentials.json`; the **staker / subscription tiers** unlock self-learning skill detection, the memory linter, and remove the local cap; if you need those features the OSS substrate is clean MIT but the activated path Evaluation: `pip install sibyl-memory-cli && sibyl init` is the install path; for the Hermes Agent integration: `pip install sibyl-memory-hermes && sibyl-memory-hermes install-plugin`; for direct SDK use: `pip install sibyl-memory-client`; for MCP exposure: `pip install sibyl-memory-mcp`; the free unactivated tier is the right starting surface

Who should skip it

Skip Sibyl-Labs/Sibyl-Memory unless the captured evidence suggests it solves a problem you are actively working on.

About this signal

Sibyl-Labs/Sibyl-Memory is tracked by RepoRadar as an AI project in the MIT local-first agent-memory family section. First seen 2026-06-25; the source record was last checked on 2026-06-25. The current verdict is 'try now' with a Silver tier and easy setup difficulty. Across RepoRadar's eight signals, Sibyl-Labs/Sibyl-Memory is strongest on workflow potential (9.2) and novelty (9.0) and weakest on momentum (7.0) — a profile worth weighing against your own priorities. This page summarizes the public evidence on the linked source page and states where additional review is still needed.

How this item is evaluated

The Sibyl-Labs/Sibyl-Memory record combines a 7.7/10 composite score with separate popularity (88.0), risk (conditional), and setup (easy) signals. See the scoring methodology for the current weights and evidence definitions.

Putting this into practice? Read How to vet an AI agent or MCP server before you wire it in for the checklist behind this score.

Risk explanation

free unactivated path makes no network calls; activated path binds a wallet or email to ~/.sibyl-memory/credentials.json and uses a vendor-controlled tier-check endpoint at api.sibyllabs.org; staker/subscription tiers required for self-learning skill detection; memory linter.

Evidence links
Closest alternatives / related signals
sibyl-memory sibyl-labs local-first sqlite fts5 no-vectors no-embeddings no-retrieval-model
Verification record

What RepoRadar actually verified

Tested in a bounded workflow

Bounded representative workflow retained by RepoRadar verification harness. Last checked 2026-07-13T10:37:20.247588Z.

passed · cohort-20260712-sibyl-local-memory-workflow

Tester
RepoRadar automated local verification harness
Started
2026-07-13T10:37:19.526071Z
Completed
2026-07-13T10:37:20.247588Z
Environment
Windows 10 AMD64; Python 3.11.9; credential-stripped child environment; disposable home/cache
Install/setup time
1 minute(s)
Evidence scope
Bounded representative workflow
Cleanup
Per-check temporary home and work directory removed. Shared cohort package cache removed.
Actions exercised
  • Created a disposable home, work directory, and isolated package cache with credential-like environment variables excluded.
  • Created 1 synthetic fixture file(s) inside the disposable work directory; retained hashes prove the exact inputs.
  • Opened Sibyl's work-local SQLite client, stored two structured runbook entities, and wrote one acted journal event.
  • Retrieved checkout by category/name, searched the FTS index, asserted its body and hit, and retained database plus JSON proof hashes.
  • Executed bounded check: Write, retrieve, journal, and search synthetic runbook memory through Sibyl's local SQLite client.
  • Captured the complete sanitized stdout, stderr, exit status, artifact checks, and 0.72-second wall time.
Observed results
  • Command exited 0 after 0.72 seconds.
  • Sibyl persisted both runbooks, returned the structured checkout entity, found it by text search, and recorded a journal event.
  • Expected marker 'CHECK_OK entity=checkout hits=1 journal=written' was observed in retained output.
  • Validated sibyl.db: 0 required marker(s) present and 0 excluded marker(s) absent; size and SHA-256 are retained.
  • Validated result.json: 4 required marker(s) present and 0 excluded marker(s) absent; size and SHA-256 are retained.
Observed strengths
  • The local client combined structured entity CRUD, an append-only journal, and FTS lookup in a single embedded SQLite file.
Friction
  • The local search path depends on SQLite FTS5 support and the client uses generated identifiers and timestamps that vary per run.
  • Setup or runtime emitted 1 stderr line(s); the complete warnings/errors are preserved in the retained log.
Limitations
  • The local free-tier path validates entity and journal persistence plus FTS search, not account activation, self-learning, synchronization, paid tiers, or remote APIs.
  • This credential-free disposable workflow does not establish production scale, model quality, reliability under sustained use, or team adoption.

Pricing assessment: The unactivated local client stayed on its embedded free path and used no Sibyl account, sync service, model, or paid feature.

Privacy assessment: Synthetic runbooks and the journal entry were written only to the disposable SQLite database, which the harness removed after hashing.

Open retained test log →

Verification sources

Longitudinal intelligence

How this decision record is moving

Raw history JSON →

40 dated snapshots retained from 2026-06-25 through 2026-08-13; see the snapshot index for explicit coverage gaps. Stars, version, release, pricing, integration, risk, maintenance, verdict, score, and momentum fields remain explicit even when a source has not reported them. Repository momentum is a normalized 0–10 RepoRadar signal; GitHub stars appear only where the popularity monitor retained exact timestamped observations.

RepoRadar score7.7 current · +0.0 net
Repository momentum9.0 current · +2.0 net
GitHub stars (observed)99 current · +4 net
GitHub stars99 exact observation
Versionv0.1.0
Last release2026-05-21T00:02:26Z
Maintenanceactive
Current riskconditional
Current verdicttry now
Pricing baselineNo structured commercial pricing baseline
Pricing checkedNot applicable or not recorded
Pricing freshnessNo dated commercial pricing review
Integrations baselineClaude, Model Context Protocol

Recent dated points

DateScoreMomentumStarsRiskVerdictMaintenance
2026-08-137.79.099conditionaltry nowactive
2026-08-127.79.099conditionaltry nowactive
2026-08-117.79.098conditionaltry nowactive
2026-08-107.79.098conditionaltry nowactive
2026-08-097.79.098conditionaltry nowactive
2026-08-087.79.098conditionaltry nowactive
2026-08-077.79.098conditionaltry nowactive
2026-08-067.77.0Not recordedconditionaltry nownot recorded
2026-08-057.77.0Not recordedconditionaltry nownot recorded
2026-08-047.79.098conditionaltry nowactive
2026-08-037.79.098conditionaltry nowactive
2026-08-027.79.098conditionaltry nowactive

Why the record changed

stars changed

Stars changed: 98 → 99.

stars changed

Source-observed stars changed: 97 → 98. This reports the retained observation delta and does not infer why the upstream change occurred.

stars changed

Stars changed: 95 → 97.

verification changed

Verification changed: Discovered → Tested in a bounded workflow.