TekmerDB

The database that knows when it's wrong.

TekmerDB gives AI agents reliable memory — storing not just facts, but how confident to be in each one, where each came from, and when two sources disagree. Unlike a vector database that retrieves everything with equal confidence, TekmerDB detects contradictions mechanically, tracks source reliability over time, and tells you what it doesn't know.

Tekmer comes from the Turkish word for singular, unique, one-of-a-kind — one storage layer that does what no other database does: reason about the reliability of what it holds.

⚠️ MVP — Pre-release software. TekmerDB is under active development. APIs may change and it has not been audited for production use.

---

Navigation

• Who this is for
• What it is. What it isn't.
• How is it different from a vector database?
• Install
• Quick start
• See it in 30 seconds
• Benchmark: TekmerDB vs RAG
• MCP — connect any AI agent
• Learn more
• License
• 📖 Full Wiki — deep documentation

---

Who this is for

AI engineers and developers — drop TekmerDB behind your RAG pipeline. Your agent gets a memory layer that knows when two sources disagree instead of returning both with equal confidence.

Tech leads and architects — evaluating reliability infrastructure for AI systems? Start with The Fundamental Design Philosophy and Why TekmerDB Is Not A Truth Machine.

Business and compliance — deploying AI in a regulated industry (energy, healthcare, legal, finance)? TekmerDB is the audit trail and provenance layer that makes AI decisions traceable. Read Why Conflicts Are More Valuable Than Clean Answers.

---

What it is. What it isn't.

TekmerDB is	TekmerDB is not
A reliability layer for AI agent memory	A general-purpose database
A mechanical engine for epistemic trust	A truth machine
An audit trail for AI decisions	A replacement for your application logic
EU AI Act compliance infrastructure	A vector database with extra features
Air-gapped, no cloud, no API keys	A hosted SaaS

---

How is it different from a vector database?

A vector database finds similar things fast. It has no concept of whether those things are reliable.

	Vector DB (RAG)	TekmerDB
Storage unit	Text chunk	Probabilistic Fact Object (PFO)
Confidence	None — all results equal	Mechanically computed 0.0–1.0
Contradiction detection	None	NLI classifier on every insert
Source tracking	Filename + page	UUID, weight, full corroboration history
Poisoned data	Returned as fact	Flagged, source degraded, conflict named
Audit trail	None	Full provenance to EU AI Act standard
Deployment	Cloud-dependent	Single air-gapped binary

TekmerDB is additive, not a replacement. Pipe the outputs of your existing RAG pipeline into TekmerDB and your agent gains a memory layer that knows what to trust.

---

Install

Prerequisites: Linux x86_64, wget

# 1. clone the repo
git clone https://github.com/raa82/tekmerdb
cd tekmerdb

# 2. run the installer
sudo ./install.sh

The installer will:

• Download the pre-built binaries from the GitHub release
• Download the ML models (~420 MB total) from HuggingFace
• Install everything to /opt/tekmerdb
• Copy the default config file

Start the engine:

cd /opt/tekmerdb && ./tekmerdb

The engine listens on http://127.0.0.1:3000 by default.

Configure:

nano /opt/tekmerdb/tekmerdb-server.conf

---

Quick start

# insert a fact
curl -X POST http://localhost:3000/pfo \
  -H "Content-Type: application/json" \
  -d '{
    "claim_text": "North Sea wind capacity reached 35 GW in 2024",
    "confidence": 0.8,
    "source": "IEA Energy Report",
    "domain": "CriticalInfrastructure"
  }'

# semantic search
curl "http://localhost:3000/search?q=North+Sea+wind+capacity&k=5"

# check source reliability
curl "http://localhost:3000/source?name=IEA+Energy+Report"

---

See it in 30 seconds

Insert two contradicting facts. Watch what happens.

# Fact from a trusted source
curl -s -X POST http://localhost:3000/pfo \
  -H "Content-Type: application/json" \
  -d '{
    "claim_text": "Global coal demand will fall 20% by 2035",
    "confidence": 0.8,
    "source": "IEA World Energy Outlook",
    "domain": "CriticalInfrastructure"
  }'

# Contradicting claim from a lobby group
curl -s -X POST http://localhost:3000/pfo \
  -H "Content-Type: application/json" \
  -d '{
    "claim_text": "Global coal demand will increase 40% by 2035",
    "confidence": 0.8,
    "source": "CoalIndustryLobby2024",
    "domain": "CriticalInfrastructure"
  }'

Both facts flagged. Both confidences reduced. Conflict preserved. Source named.

{
  "id": "a1b2c3...",
  "confidence": 0.52,
  "conflict_refs": ["e5f6g7..."],
  "corroboration_count": 0,
  "source": "CoalIndustryLobby2024"
}

A vector database returns both claims with identical authority. TekmerDB flags the conflict, names the source, reduces confidence on both, and preserves the full provenance chain. No hallucination. No silent resolution.

---

Benchmark: TekmerDB vs RAG

9 compliance questions. Same LLM. Real 510-page IEA document. Three industry sources.

Test	Question	Winner
1	Global energy demand by 2035	TekmerDB — 7 conflicts flagged, RAG blended contradictions silently
2	1.5°C climate target	TekmerDB — contradictions detected, confidence 0.72
3	EU AI Act certification	TekmerDB — clear NO with reasons, RAG returned irrelevant data
4	Poisoned data (lobby claim)	TekmerDB — fake claim flagged and source named. RAG opened with it as fact.
5	Source audit trail	Tie
6	Regulatory submission decision	TekmerDB — compliance verdict with confidence score and source trace
7	2024 actual energy demand	TekmerDB — correct source retrieved, RAG returned projections
8	Oil demand 2035 and 2050	TekmerDB — 2 conflicts flagged correctly
9	Fastest growing energy sources	Tie

Final score: TekmerDB 7 — RAG 0 — Ties 2

Read the full comparison report →

---

MCP — connect any AI agent

TekmerDB ships with a Model Context Protocol server. Connect any MCP-compatible agent directly to the engine.

Claude Desktop config:

{
  "mcpServers": {
    "tekmerdb": {
      "command": "/opt/tekmerdb/tekmerdb-mcp"
    }
  }
}

The agent can insert facts, search, check source reliability, and update confidence — all through natural language. No API key. No cloud.

---

Learn more

• Why Traditional RAG Fails — the core problem TekmerDB solves
• The Confidence Model — how confidence is computed mechanically
• Conflict Detection — why contradictions are preserved, not resolved
• The Epistemic Model — the philosophy behind the architecture
• Full Wiki — complete documentation

---

License

Apache 2.0 — see LICENSE.

Enterprise features (audit log exports, RBAC, EU AI Act compliance reporting, managed hosting) are available under a commercial licence. Contact us.