Back to Agents That Account for Themselves
working paper2026Runtime knowledge graph paper

Live Knowledge Graphs: Runtime Graph Integration for Continuous Domain Adaptation in Language Agents

Live Knowledge Graphs argues that knowledge graphs should remain active at inference time. Instead of distilling a graph into weights and discarding it, a language agent can query a live graph, receive a provenance-complete context slice, and answer against current evidence.

Paper workspace

Live draft structure

working-draft

Artifacts

Markdown paper source

Runtime knowledge graph paper mapped from the local source manuscript.

source-only

Editable source

Markdown paper source exists. It should converge with the Graph Kernel paper without duplicating claims.

Source anchors

live-knowledge-graphs/paper/paper.md

Comp-Core/papers/runtime-knowledge-graphs/paper.md

Comp-Core cc-graph-kernel service

Method tags

runtime graph integrationcontext slicingadmissibility tokensprovenance

Ingest intersections

knowledge-graphruntime-raggraph-kernelprovenance

Status

Paper drafted; cc-graph-kernel production service is the implementation substrate.

Key claims

01

Training-time graph integration goes stale in long-running agent systems.

02

Runtime context slicing preserves currency and auditability.

03

Admissibility tokens make graph-grounded context verifiable.

Public reading note

Paper drafted; public reader page safe.

Standard skeleton

What this paper must keep proving

Schema

problem

Training-time graph integration goes stale when an agent's domain changes daily.

method

Query the graph live during inference and inject provenance-complete context slices rather than static summaries.

implementation

Rust cc-graph-kernel, context slicer, graph traversal endpoints, admissibility token verification.

data

Production graph triples and path-evaluation snapshots; public release should expose aggregate metrics, not private graph contents.

evaluation

Path ranking, context-slice integrity, latency, provenance completeness, and hard-negative discrimination.

references

GraphRAG, KAPING, DSS/knowledge-graph curricula, retrieval-augmented generation.

openQuestions

Which parts should be merged into Graph Kernel v2 versus remaining a companion paper.

Checkpoints and references

Proof chain

paperpending

Claim checkpoint

central-claim slot

Every central claim must point to a proof anchor or remain labeled as speculative.

implementationpending

Implementation checkpoint

implementation-map slot

Every method should identify the code path, harness, schema, or protocol that embodies it.

experimentpending

Evidence checkpoint

evidence-manifest slot

Every reported result should point to run IDs, packet IDs, data snapshots, commits, or review artifacts.

external-referencepending

Reference checkpoint

references slot

Every external claim should resolve to a cited paper, benchmark, standard, or documented prior system.

paperpending

Release checkpoint

release-gate slot

Every PDF needs a named condition before it can move from draft to citation-ready.