index code

The most secure AI agent platform.

1 / 4

intranet · 0 egress

➜ intranet idx team run "audit Q1 commits for security issues"

┌─ spawn

│ agent-1 code-review → deepseek-coder-v3

│ agent-2 threat-model → glm-4-plus

│ agent-3 report-writer → qwen2.5-72b

└─ 3 agents · isolated VPC · 0 external calls

┌─ run

│ agent-1 scanning 847 commits across 34 modules

│ agent-2 threat modeling · OWASP Top 10 + CWE Top 25

│ agent-3 drafting executive report

└─ completed in 4m 12s · all artifacts local

┌─ findings

│ 7 critical auth bypass · SQLi · XSS

│ 14 medium missing input validation

│ 23 info code quality suggestions

└─ report saved to .idx/reports/2026-Q1-audit.pdf

✓ all data stayed on network · 0 bytes egressed

➜ intranet idx team ship "add 2FA to login flow"

┌─ team · feature-delivery

│ planner deepseek-v3

│ coder deepseek-coder-v3

│ tester qwen2.5-coder-32b

│ reviewer glm-4-plus

└─ 4 agents · isolated sandbox

┌─ plan

│ 6 files to change · 18 tests to add

│ estimated 4m runtime

└─ approved

┌─ execute

│ ✓ auth/totp.ts +142 −0

│ ✓ auth/login.ts +18 −4

│ ✓ middleware/2fa-guard.ts +74 −0

│ ✓ tests/2fa.spec.ts +218 −0 · 18 cases

│ ✓ migrations/add-2fa-secret.sql +12 −0

└─ applied

┌─ verify

│ tests 18 / 18 passed

│ typecheck clean

│ review approved · no blockers

└─ ready

✓ PR draft at .idx/pr/add-2fa.diff

➜ intranet idx dept create "customer-success" --interactive

? what should this department do?

handle tier-1 customer tickets end-to-end

? models to use?

[x] deepseek-v3 reasoning & routing

[x] qwen2.5-72b drafting replies

[x] glm-4-plus intent classification

[ ] llama-3-70b

┌─ provisioning on intranet-cluster

│ ✓ 3 agents assembled

│ ✓ connected to internal Zendesk

│ ✓ shared memory · encrypted at rest

│ ✓ audit log streaming to SIEM

└─ ready

✓ department "customer-success" is live

monitor: https://internal.idx/dept/customer-success

➜ intranet idx verify --network-isolation --window 1h

┌─ outbound connections

│ DNS ✓ only intranet resolvers

│ HTTP ✓ 0 external calls

│ HTTPS ✓ 0 external calls

│ model API ✓ via local proxy only

└─ clean

┌─ data residency

│ training data ✓ all local

│ conversations ✓ encrypted at rest

│ model weights ✓ on-premise only

│ audit trail ✓ immutable · 90d retention

└─ clean

┌─ compliance posture

│ 网络安全法 ✓

│ GDPR ✓ (EU-ready)

│ SOC 2 Type II ○ audit-ready

│ ISO 27001 ○ audit-ready

└─ clean

✓ 100% intranet · 0 bytes egressed in audit window

SECURITY

Runs entirely inside your firewall.

Open-source models. Proprietary agents. Zero external calls.

SOURCE AVAILABLE
·
ON-PREM
·
ZERO EGRESS
·
AUDIT-READY

Your code never touches an external API. Your model weights stay on your hardware. Your agents log to your SIEM. Audit every line of agent code at deploy time.

ARCHITECTURE

Build AI departments that run themselves.

Each agent picks its own model. The team coordinates like a real org chart.

Engineering Ops

planner · deepseek-v3
coder · deepseek-coder-v3
tester · qwen2.5-coder-32b
reviewer · glm-4-plus

Security Ops

code-review · deepseek-coder-v3
threat-model · glm-4-plus
report-writer · qwen2.5-72b

Customer Ops

router · glm-4-plus
drafter · deepseek-v3
escalation · llama-3-70b

Engineering is where we start. Teams for sales, support, HR, and legal follow the same pattern — define the roles, pick the models, run.

MODELS

Any open-source model you can download.

Mix and match. Upgrade independently. No vendor lock.

Browse all models on HuggingFace →

BENCHMARKS

Your model choice drives agent capability.

index code runs any open-source model. Pick the one that scores best on your workload.

See how open models rank on SWE-Bench →

index code itself is not on any public leaderboard — it's evaluated on each customer's private datasets during pilot.

HOW IT WORKS

Three steps to production.

01 · DEPLOY

Install the index code control plane on your intranet. One command. No external dependencies.

02 · CONNECT MODELS

Point index code at your open-source model servers (local vLLM, Ollama, or your existing inference cluster).

03 · LAUNCH TEAMS

Define agent teams in YAML or through the UI. Watch them ship work while you sleep.

GET STARTED

Ready to deploy AI agents inside your enterprise?

30-minute demo. No sales pressure. Your infrastructure, your rules.