∙v 0.1.7
  EXP
lat 31.22°N · lon 121.47°E · uplink ok
cpu 12%mem 4.2gb 02:18:44 UTC
▸ investraness · Harness · v 0.1.7

Evolution ·
for Investment intelligence.

AI Powered Engineering Velocity

3,101
graph nodes
16,977
edges
82%
token saved
17ms
p50 retrieval
scroll to enter
▣ 01 · interactive data harness

Agentic AI
for
Investment-Analysis

Natural language queries grounded in your codebase, deployments and incident history. Every answer traces back to its sources — no hallucination, no black boxes.

◆ grounded in 18 retrieved nodes
◆ auto-citations on every claim
◆ p95 response · 1.4s · streaming
Agent LIVE
Plan 5
Graph RAG 284K
Knowledge 12
ready

Agent run · o3-mini-high

The agent is parsing the editor text, decomposing it into subtasks, and consulting Graph RAG + Knowledge for grounded context.

Parse input markdown
step 1 · 124ms · tokens 482
Detected 6 sections, 4 math blocks, 1 image, 3 inline annotations.
Retrieve context from Graph RAG
step 2 · 312ms · 1,240 nodes traversed
Pulled 8 community summaries centered on investraness.
Annotate inline references
step 3 · 87ms · 5 spans tagged
2 blue3 amber queued for review.
Render preview · drafting
step 4 · streaming · 1,240 / 3,800 tokens
LaTeX blocks, image embed and annotation chips being applied.
Persist run to history
step 5 · pending

Tool calls

graph_rag.search("gradient descent non-convex", top_k=8) knowledge.match(annotations=[2,5], mode="hybrid") editor.apply_inline(class="ann-inline", spans=[…])

Plan · current run

Decomposition of the input into 5 ordered subtasks. ETA is computed from observed p50 latency over the last 50 runs.

T1
Parse & structure markdown
tokenizer · 482 tokens
~0.2s
T2
Retrieve graph neighborhood
Graph RAG · depth 2 · 1,240 nodes
~0.4s
T3
Resolve inline annotations
Knowledge · 5 candidates
~0.3s
T4
Render preview (math + image)
KaTeX · 1 plot embed
~0.6s
T5
Persist run metadata
history · jsonl append
~0.1s

Total ETA

expected
1.6s
p95
2.4s
parallelism
T1 → T2 ‖ T3 → T4 → T5
cost est.
0.0042 USD

Graph RAG · 284,231 nodes 1.2M edges 512 communities

Subgraph retrieved for the active query. Hover any node to expand its 1-hop neighborhood.

gradient descent loss-fn non-convex saddle minima adam sgd lr momentum

Top relations

gradient→ USES →adam, sgd
descent→ APPLIED_TO →non-convex
non-convex→ HAS_LOCAL →saddle, minima
loss-fn→ DIFFERENTIABLE →gradient

Knowledge · top matches

Documents matched against the current input, ranked by hybrid score (semantic + keyword + graph proximity).

MD
gradient-descent.md
notes/optimization/ · 1.4kb · hybrid 0.94
L1–L18
PY
sgd_optimizer.py
src/optim/ · 2.1kb · hybrid 0.89
42 ln
NB
convexity.ipynb
notebooks/ · 18kb · hybrid 0.81
24 cell
PDF
boyd_convex_opt.pdf
refs/ · 2.3mb · hybrid 0.76
p. 1–42
MD
learning-rate-schedule.md
notes/ · 0.6kb · hybrid 0.71
L1–L9
PY
adam.py
src/optim/ · 1.2kb · hybrid 0.68
28 ln
82%
tokens saved by DeepSeek MLA
▣ 02 · TOSA · token system attention

Spend less.
Think more.

MLA(Multi-head Latent Attention)compresses prompts, realized context caching — cutting token cost without dimming the lights on answer quality.

semantic cache hit
71.4%
spend · 30d
$ 12.4k − 82%
▣ 03 · knowledge base

Time-Varying Persistence
memory.

Every doc, runbook, ADR, post-mortem and chat thread — versioned, embedded, and queryable in milliseconds.

DOC

RAG-Plugin

Markdown, pdf, Json and Docs ingested with structure preserved.

indexed14,284
CDE

Codebase

Repo-aware AST chunking. Symbols, comments and tests linked to services.

repos62 · 1.4M loc
RTM

Runtime

Live execution context — container state, resource usage, and call stacks captured on the fly for grounding agents in real-time.

active312
SD

Sand-box

Isolated playground where agents prototype and verify before production.

archived189
ORC

Orchestrator

Smart router — picks the right retriever, tool and model per query.

messages412k
Vet

Vector

Dense embeddings store with sub-second semantic recall.

schemas284
02:18:14+ADD  runbook · auth-fallback-v2.md2.4 kb
02:17:58~UPD  repo · payments-gateway @ a7c91f4+148 / −22
02:17:32+ADD  postmortem · 2389 · circuit breaker timeout8.1 kb
02:16:01~UPD  openapi · checkout-svc v2.43 routes changed
02:14:44−DEL  deprecated · legacy-auth runbookarchived
▣ 04 · graph rag · exp

Query the
dynamic graph.

Datasets, events, services, people and docs woven into a single semantic mesh. Retrieval grounded in topology, not just text similarity.

Explore the
Portfolio Optimization.

Built for researchers who refuse to wait for the future.

▸ initialize console ▸ Dashboard
· interactive data ·· TOSA ·· knowledge base ·· graph rag ·· pipelines ·
//Authentic Evolution · technology for the future · 2026
▣ 05 · access · self registration

Claim your
access.

Your account is created instantly and attached to the default project. The key is shown in plaintext only once — save it immediately.

◆ Instant account · Auto-attached to the default project
◆ Default key rate limit 60 RPM / 100K TPM
◆ Connect to LLMs via the router · Multiple models to choose from
API ENDPOINT · http://115.190.165.230:8080
REGISTER · NEW USER SIGN-UP
Already have an account? Sign in to the console →