Veyra

Persistent engineering memory & Knowledge Observatory for DeepSeek Harness.

Install it, restart DSH, keep working. Veyra observes useful context, remembers durable lessons, indexes RAG documentation, and recalls them via unified Hybrid Search. It does not write into your repo.

Install

dsh plugin --profile web add @lovedolove/veyra

Restart dsh web. A good boot looks like this:

[veyra] plugin loaded (home=...)
[veyra] registered /veyra
[veyra] registered skill: veyra
[veyra] registered tools: veyra_remember, veyra_recall, veyra_inspect, veyra_forget, veyra_promote

Requires Node 22.5+ and DSH >=0.1.2-rc.1. Memory lives in $DSH_HOME/veyra/.

Daily use

You do not manage memory for ordinary work.

1. Do engineering in DSH as usual.
2. Veyra distills the turn into a candidate (files, symbols, test outcomes, claim signal, and structured causal facets: symptom → rootCause → remedy → verifiedOutcome). Durable, grounded lessons become derived memory and are linked to neighbors — extended, updated, or contradicted. Near-duplicates are not cloned; instead, repeated observations and verified test outcomes strengthen existing knowledge (unverified → reviewed → verified) and accumulate evidence anchors. Nothing is merged silently.
3. On a later session in the same project, relevant memory and RAG knowledge are recalled automatically via Hybrid Search (combining lexical FTS5 BM25 + exact symbols, token-level semantic Jaccard overlap, intent affinity, and relationship graph signals). Contradictions stay visible on both sides. Superseded, repository-drifted, and long-idle unverified memories leave ambient recall. Canonical is never assigned automatically.

Ask the agent to remember something important, or just keep going — automatic recall is already on.

Hybrid Search & Unified RAG

Veyra unifies Engineering Memory (kind: 'memory') and Knowledge Base / RAG documentation (kind: 'knowledge'):

• Lexical: SQLite FTS5 BM25 ranking plus exact symbol and file-path boosts.
• Semantic: Token Jaccard overlap and query concept coverage.
• Memory & Causal: Intent affinity prioritizing root cause for why, remedy for how, verified outcomes for did this fix it.
• Relationships: Graph cohesion boosts interconnected knowledge.
• Multi-Dimensional Ranking: Composite score with full transparent breakdown (relevance, semantic, evidence_strength, validation_tier, scope_proximity, freshness_tier, confidence, intent_affinity, relationship).

Commands & Knowledge Observatory

Command	What it does
/veyra	Status and memory counts for this workspace
/veyra observatory	Open the Human Knowledge Observatory dashboard
/veyra observatory search <query>	Inspect hybrid search signals and score breakdowns
/veyra observatory record <id>	Deep inspection answering the 6 questions (What, Why, Where, When, Evidence, Validation)
/veyra observatory causality	Causal knowledge map (symptom → rootCause → remedy → verifiedOutcome)
/veyra observatory relationships [id]	Directed graph projection of knowledge links
/veyra observatory local <id>	1-hop Local Graph around a record (default graph view)
/veyra observatory graph [id]	Local Graph when an id is given; otherwise the project edge list
Network Graph WebUI	Host page at /veyra?id=<record> (Local Graph, 1 hop default). Search → record → graph
/veyra observatory contradictions	Active conflicting claims shown side-by-side
/veyra recall <query>	Hybrid search project (+ reusable) memory
/veyra recent	Last 8 records, including candidates
/veyra inspect <id>	Read one record with deep provenance and causal facets
/veyra forget <id>	Soft-forget (leaves recall, stays inspectable)
/veyra promote <id> canonical	Mark as project truth — only this is canonical

Tools with the same names (veyra_remember, veyra_recall, veyra_inspect, veyra_forget, veyra_promote) are available to the agent. Load the bundled veyra skill when you need full guidance.

Memory rules

What	Authority	Auto-recalled?
Auto-captured observation	candidate	no
Remembered / learned lesson	derived	yes
Explicitly promoted truth	canonical	yes

Canonical is never assigned automatically. Repo files stay authoritative. Similarity is not identity: overlapping memories are linked, not merged. Secrets are redacted. Projects are isolated; reusable memory is opt-in.

Config (optional)

In the profile cordis.patch.yml:

- id: veyra
  name: '@lovedolove/veyra'
  config:
    recallLimit: 5
    includeReusable: true
    observe: true
    learn: true

VEYRA_HOME overrides the storage root.

License

MIT