Skip to main content

​
Clockchain

Temporal causal graph for AI agents. PostgreSQL-backed directed graph of historical moments — canonical spatiotemporal URLs, typed causal edges, autonomous expansion, and browse/search/discovery APIs. Currently holds 1,900+ nodes and 5M+ edges spanning 700 BCE to 2026, with 73% image coverage and growing.

GitHub

timepointai/timepoint-clockchain — Apache-2.0, Python 3.11+, FastAPI

Detailed Docs

Full API reference, graph architecture, and deployment docs

​
MCP Endpoint

The Clockchain exposes a Model Context Protocol (MCP) interface for AI agents and tools:
PropertyValue
URLhttps://clockchain.timepointai.com/mcp/
Versionv1.26.0
TransportStreamable HTTP
AuthPublic reads (no auth required)
Any MCP-compatible client (Claude Desktop, Cursor, etc.) can connect directly to browse and search the temporal graph.

​
Why This Exists

AI agents that reason about causality across time currently rely on web search (noisy, unstructured), knowledge graphs (no temporal dimension), or hallucination. The Clockchain is a structured alternative: every node carries dialog, entity states, provenance, and confidence, addressed by a canonical spatiotemporal URL, in a format (TDF) designed for machine consumption.

​
Graph Architecture

Two PostgreSQL tables: nodes (canonical spatiotemporal URLs as PKs) and edges (source, target, type, weight).

​
Canonical URL Format

/-44/march/15/1030/italy/lazio/rome/assassination-of-julius-caesar
8 segments: year (negative = BCE), month, day, time (24hr), country, region, city, slug.

​
Content Layers

LayerContentSource
0URL path + event nameAuto-generated
1Metadata: figures, tags, descriptionExpander (LLM)
2Full Flash scene with dialog, characters, image, CDN-hosted image_urlFlash renderer

​
Edge Types

TypeMeaningAuto-linked?
causesDirect causal relationshipNo — expander or manual
contemporaneousSame year (+/- 1)Yes, weight 0.5
same_locationMatching geographyYes, weight 0.5
thematicOverlapping tagsYes, weight 0.3

​
Background Workers

WorkerRole
RendererCalls Flash to upgrade layer-1 nodes to layer-2 (full scenes)
ExpanderLLM-driven graph growth — discovers related moments, renders each through Flash
JudgeContent moderation
Daily”Today in History” cron — adds new nodes for the current date

​
Public API

No authentication required for read access: 
# Graph statistics
GET /api/v1/stats

# List moments (paginated, filterable)
GET /api/v1/moments?limit=10&offset=0&q=rome&year_from=-500&year_to=500

# Get a specific moment
GET /api/v1/moments/{canonical-path}

​
Rate Limits

TierLimit
Public (unauthenticated)60/min
Authenticated reads300/min
Authenticated writes30/min

​
The Bayesian Flywheel

Each new event with causal edges tightens the Bayesian prior — fewer plausible things could have happened in the gaps — approaching asymptotic coverage of any historical period. Flash scenes and Pro simulations feed into the Clockchain graph, which builds a stronger Bayesian prior. That prior drives the Expander (LLM) to discover gaps, which feeds back into more Flash and Pro renders. 
Flash scenes ──► Clockchain Graph ──► Stronger Prior
Pro simulations ─┘        │               │
                          ▼               │
                    Expander (LLM) ◄──────┘
                    discovers gaps
The name is conceptual. This is PostgreSQL, not a blockchain.