Vestige
<div align="center">
# Vestige
### Memory and a safety gate for AI coding agents.
</div>
---
## What it does
- **Remembers.** Agents keep what happened in earlier sessions and can trace a failure back to the change that caused it.
- **Stops bad tool calls.** 102 fixed rules check every tool call before it runs and block things like recursive deletes and credential dumps.
- **Asks again.** An earlier yes is not a yes now. A one-time owner permit lasts 15 minutes and then the agent has to ask again.
- **Proves it.** Every write is signed and hash chained, so anyone can check the log offline.
It runs on your machine on a plain CPU. No LLM, no GPU, no cloud calls.
---
## Quick start
```bash
npx @vestige/init --guard
vestige serve
```
---
## Benchmarks
These measure retrieval, meaning whether the right memory came back in the top 5. They do not measure answer accuracy. Every number is evaluated on 100% of the benchmark questions with zero embeddings, zero vectors, and zero cloud LLM calls ($0.00 spend).
| Benchmark | Scale & Setting | Vestige v6.2.1 | Plain BM25 |
| :--- | :--- | :--- | :--- |
| **LongMemEval_S** | Official all gold sessions in top 5 (470 non-abs) | **91.28%** | 82.98% |
| **LongMemEval_S** | Any gold session in top 5 (all 500 questions) | **98.60%** | 97.00% |
| **LongMemEval_M** | Scaled haystack (~476 sessions, ~4,894 turns/Q; 470 non-abs) | **91.70%** Any@5 / **76.38%** All@5 | — |
| **LoCoMo-10** | Full conversational dialogues, right session in top 5 (1,536 CORE Qs) | **92.25%** | 87.24% |
| **LoCoMo-10** | Conversational turn-level soft recall in top 5 (1,536 CORE Qs) | **78.13%** | 69.99% |
| **MemoryAgentBench** | Multi-hop reasoning, 421K distractor context (`ruler_qa2`, 100 Qs) | **98.00%** Any@5 / **93.00%** All@10 | — |
| **MemoryAgentBench** | Needle retrieval in 197K context (`ruler_qa1`, 1,204 docs, 100 Qs) | **95.00%** R@5 / **99.00%** R@10 | — |
| **MemoryAgentBench** | Merged 400K scenarios (`longmemeval_s*`, 5 sets, 287 non-abs) | **94.08%** Any@5 / **85.02%** All@5 | — |
| **BEAM 100K** | 20 conversations, 2,127 sub-sessions, right session in top 5 (355 Qs) | **66.20%** Any@5 / **74.93%** Any@10 | — |
| **BEAM 100K** | Turn exact ID recall in top 5 (355 scored questions) | **45.92%** Any@5 / **55.77%** Any@10 | — |
This is the top retrieval result without embeddings we know of on LongMemEval_S. Settings were tuned on these datasets. As a blind split check, LoCoMo held-out conversations 6 to 10 score **91.75%** session recall at 5 and **76.03%** turn soft recall at 5.
On a plain CPU with 8 GB of RAM, a LongMemEval question takes ~500 ms, a LoCoMo question ~7 ms, and a 5,000-turn LongMemEval_M haystack ~5 seconds, with peak RSS under 350 MB.
Full tables, category breakdowns, and zero-leak disclosures are in [benchmarks/README.md](benchmarks/README.md).
```bash
bash benchmarks/fetch_data.sh && bash benchmarks/run_all.sh
```
---
## Learn more
- [Getting started](docs/GETTING-STARTED.md)
- [Safety gate rules](operator/README.md)
- [Storage and signed log](docs/STORAGE.md)
- [Tool reference](docs/REFERENCE.md)
- [FAQ](docs/FAQ.md)
---
Built in Rust by Sam Valladares. Licensed under AGPL-3.0.
TDQS
Scored across 34 tools
Most tools have distinct purposes: search, memory, codebase, intention, and various merge/explore steps are clearly separated. However, cross_reference is an explicit alias for deep_reference, creating redundancy, and the cluster of merge-related tools (find_duplicates, merge_candidates, plan_merge, apply_plan, merge_undo) could be confused despite having distinct stages.
Naming is mixed: many tools follow a verb_noun pattern (smart_ingest, find_duplicates, plan_merge, apply_plan, explore_connections, suppress), but others are noun_verb (memory_timeline, system_status, importance_score, merge_candidates, memory_health). A few are single-word nouns (search, memory, backup, export, gc, dream). This inconsistency in convention makes the set less predictable.
With 34 tools, this server far exceeds the 25-tool heavy threshold. While the domain is broad (memory management, codebase, intentions, ingestion, merging, health), many tools could be consolidated — e.g., five separate merge-related tools (find_duplicates, merge_candidates, plan_merge, apply_plan, merge_undo) and multiple health/status tools. The count feels bloated for agents to navigate efficiently.
The tool set covers the full memory lifecycle: creation (smart_ingest, memory set), retrieval (search, get, memory_timeline), update (edit, promote/demote), deletion (purge, gc, suppress), plus backup/export/restore and exploratory analysis. Minor gaps exist, such as a dedicated bulk delete (though gc covers it) and no explicit 'list all memories' without time constraints, but these are workable.