Decision recorded
Reason forgotten.
Open Core · Apache 2.0
Organizations rarely lose information. They lose relationships between information.
Retineo reconstructs hidden relationships across documents, decisions, conversations, and code so engineering teams recover context search never finds.
Not a plugin. Not a chatbot. A compiler for engineering context.
Apache 2.0
Open Core
Local-first
CLI · REST · MCP
HNSW + BM25
Search finds documents. Your team needs understanding.
Every artifact exists somewhere. None of them point to each other.
Incident documented
Lessons disconnected.
Architecture exists
Tradeoffs disappeared.
Documentation survived
Understanding didn't.
Search retrieves evidence. Retineo reconstructs relationships.
Same query. Two completely different answers.
Search: Redis
redis-config.yml
incident-2024-03.md
architecture-2022.png
slack-thread-checkout.json
...and 16 more files
Retineo: Redis
│ Incident #42: checkout latency
│ ADR-17: cache layer decision
│ Decision: Redis over Postgres
│ Implementation: PR #4821
│ Runbook: failover procedure
└ Current architecture dependency
Real examples
What a query returns — and what the organization actually needs.
A service outage
Search returns
47 Slack threads, 3 incident docs, 7 PRs mentioning "checkout."
Retineo reconstructs
Incident → deploy that changed rate limiter → ADR that chose Redis over Postgres → runbook written by the engineer who left.
Onboarding a new engineer
Search returns
Onboarding checklist, architecture diagram, team wiki.
Retineo reconstructs
Onboarding question → three ADRs → the two people who made those decisions → meeting notes where scope changed.
Understanding why Redis exists
Search returns
Every file containing "Redis."
Retineo reconstructs
Performance requirement from 2022 → RFC → rejected alternative → implementation PR → current SLO.
Finding duplicated research
Search returns
Two research docs with different titles.
Retineo reconstructs
Same customer pain referenced in Q3 research, Q4 roadmap, and cancelled Q2 experiment.
What Retineo reconstructs
It does not index keywords. It traces connections between decisions, incidents, code, and documentation.
One decision and its related artifacts
Decision │ ├── ADR-17 │ │ │ └── Meeting notes (2024-03) │ ├── Incident #42 │ │ │ └── Runbook │ ├── PR #4821 │ │ │ └── Tests │ └── Architecture diagram
Decision→Incident
Meeting→ADR
Customer feedback→Roadmap
Requirement→Implementation
Issue→PR
Specification→Tests
Research→Strategy
Architecture→Deployment
How it works
Retineo compiles engineering context instead of chunking documents.
Sources
ADR, Slack, Notion, GitHub, Confluence, meeting notes, diagrams
Compiler
Normalize, structure, and extract claims into immutable artifacts
Structured artifacts
Content-addressable layers: raw, outline, essence, index
Discovery Engine
Detect references, contradictions, duplicates, and gaps
Connections
Relationship graph your team can query, browse, and verify
The open-source layers are L0 through L3. L0 ingests raw sources. L1 structures them. L2 extracts essence. L3 indexes meaning and keywords. Everything above L3 discovers the relationships between those layers.
retineo ingest README.md
$ retineo ingest README.md --verbose Source registered: README.md SHA-256: 3cd39b88... (unchanged) Status: cached Layers: L0 ✓ L1 ✓ L2 ✓ L3 ✓ Nodes: 95 total | 0 changed | 0 ghosts
Knowledge Health
Real diagnostics from a real vault. Not a mockup.
retineo health . --json
$ retineo health . --json { "score": 80, "strong": [ "good coverage", "few duplicates", "no ghost entries", "dense knowledge extraction" ], "attention": [ { "type": "orphan", "severity": "warning", "documents": [ { "contentHash": "8306dc9420d21017f5d000c66a54ad47", "sourcePath": "/home/ryzen/echo-real-test/echo-founding-member.pdf" } ], "reason": "Document has no semantic links and no inbound chunk references" }, ... 10 more orphan warnings ], "recommendations": [ "11 documents have no links or references (orphans). Consider linking them into related topics or merging." ], "advancedMetrics": [ { "metric": "fragmentation", "availableIn": "pro" }, { "metric": "contradictions", "availableIn": "pro" }, { "metric": "topicDistribution", "availableIn": "pro" } ] }
80Knowledge Score
✓ good coverage
✓ few duplicates
✓ no ghost entries
⚠ 11 orphan documents
Recommendation: Link orphan documents into existing knowledge to improve discoverability.
Generated from a real project using retineo health.
Generated from a real project using retineo health.
Open Core
Foundational layers are open source. Advanced discovery is commercial.
L0–L3
Ingest, structure, summarize, and search your own documents.
- Content-addressable storage (SHA-256)
- Structural and semantic extraction
- HNSW vector + Okapi BM25 keyword search
- CLI, REST bridge, and MCP server
L4–L9
Relationship discovery across teams, systems, and time.
- Semantic links and references
- Knowledge clusters
- Annotations and hypotheses
- Enterprise connectors and audit
Built for developers
Integrate the way you already work. No platform to adopt.
CLI
ingest · search · health
MCP
stdio JSON-RPC
REST
127.0.0.1:37891
JSON
CAS · L0–L3
Adapters
FS · JSON-RPC 2.0
Local
Ollama-first
Implementation
Concrete details from the codebase and runtime.
Storage layout
data/ ├── objects/ │ └── ab/ │ └── cdef.../ │ ├── content.md │ ├── content.meta.json │ ├── node.json │ ├── L1.md │ ├── L1.index.json │ └── L2.json ├── index/ │ ├── embeddings.jsonl │ ├── bm25.json │ ├── hnsw.bin │ ├── hnsw.manifest.json │ └── hnsw.labels.json └── retineo.sqlite
Runtime config (~/.retineo/config.yaml)
dataDir: ~/.retineo
llmProvider: ollama
embeddingModel: embeddinggemma:latest
bridge:
host: 127.0.0.1
port: 37891
search:
semantic:
topK: 100
hybridWeight: 0.7
rerank:
topK: 10
retineo status
$ retineo status RETINEO Core 0.6.4 Nodes: 95 | Sources: 95 | Pending jobs: 1 Index: 247 vectors | Last indexed: 2026-08-06T08:14:46.492Z Storage: 4.2MB | Cache: 1.8MB
Explore the project
No sales deck. Read the code and the reasoning.
Verify the architecture
Read the documentation. Inspect the source. Run the CLI.