ThreadWeave — Organizational memory that runs on your own hardware

Connects where your team already writes

No new tools. No workflow change. ThreadWeave reads from the platforms you already use and surfaces the knowledge inside them.

📧

Gmail

Google Workspace service account — polls inbox, threads replies, extracts decisions

💬

Google Chat

Listens across spaces, skips bot messages, captures human answers and decisions

📄

Google Drive

Google Docs, Sheets, Slides, PDFs, Markdown — extracts text, maps folders to teams

🔗

Microsoft 365

Graph connector push-syncs ThreadWeave entries into Copilot and Microsoft Search

# One-shot sync from Gmail (last 7 days)
threadweave gws sync --source gmail --query "newer_than:7d"

# Continuous watch mode
threadweave gws watch --interval 300

# Harvest a departing employee's knowledge
threadweave gws harvest --email lars@firm.com

The ingestion pipeline

Every piece of content passes through four stages before it reaches the knowledge store.

Dedup

SHA-256 content hash

→

Detect

Answer / decision / chat / reference

→

Filter PII

Emails, phones, IDs stripped

→

Store

Per-tenant isolation + MemPalace

# Classify text — is it worth saving?
threadweave detect "We decided to use Postgres for JSONB support"

# Output:
{
  "should_save": true,
  "content_type": "decision",
  "confidence": 0.92,
  "has_pii": false,
  "suggested_title": "Decision: Use Postgres for JSONB support"
}

Designed for people who run production systems

🏠

Self-hosted, air-gapped ready

Runs on your hardware. No cloud dependency. No external API calls required. Built for firms that can't send data off-prem.

🔐

Seven-level confidentiality

Public, internal, confidential, restricted, HR-privileged, client-confidential, legal-privileged. Auto-detected at ingest. Enforced at read.

🧹

PII-aware by default

Emails, phone numbers, personal identifiers are detected and filtered before storage. Your knowledge base stays compliant.

👋

Offboarding harvester

When someone leaves, drain their institutional knowledge — every decision, process, and client context from their email, chat, and documents.

🧠

LLM-powered classification

Uses any OpenAI-compatible endpoint to classify content with nuance. Gracefully falls back to regex heuristics when no API key is configured.

📊

Prometheus metrics

Ingest latency (p50/p95/p99), throughput, dedup hit rate, LLM vs regex fallback ratio. JSON and Prometheus text format endpoints.

🔍

Relevance-ranked search

Semantic similarity + org proximity + freshness + author authority. The right answer finds the right person.

🕐

Temporal org model

Team membership with validity windows. Query what team someone was on in Q2 2023. Knowledge graph built on MemPalace's temporal triples.

🌐

REST API + Swagger

Full OpenAPI docs at /docs. Every pipeline stage is a documented endpoint. Build your own integrations on top.

REST API that makes sense

Ingest anything. Search everything. The pipeline does the rest.

# Ingest content — pipeline handles dedup, detection, PII, storage
POST /api/v1/ingest
{
  "content": "Always run integration tests before deploying to prod.",
  "source": "manual",
  "metadata": {
    "wing": "engineering",
    "room": "deployment",
    "author_id": "harald"
  }
}

# Search across everything
POST /api/v1/search
{ "query": "deployment pipeline", "wing": "engineering" }

# Metrics in Prometheus format
GET /api/v1/metrics/prometheus

Organizational memory that runs on your own hardware