Preuve

La preuve, au centre — mesurée, sourcée et datée.

Une seule commande reproduit une réduction médiane de 69,1 % des tokens sur 7 dépôts réels (1 157 fichiers). Ci-dessous : la méthodo, les garde-fous qualité par dépôt, un estimateur de ROI piloté par la médiane en direct, un échantillon de pilote anonymisé, et une limite honnête où le benchmark agent end-to-end est encore en cours de mesure.

Preuve mesuréeindex → recherche → contexte

Reproduisez-la vous-même

Une commande, toute la preuve.

Lancez-la sur votre propre dépôt. Le même JSON alimente cette page, donc les chiffres ne peuvent jamais diverger de votre mesure.

aindex autopilot "fix billing login bug" --json
aindex pilot --repos repo1 repo2 repo3 --pdf --json
ai-indexer-mcp proof --repo ~/Documents/matchr --json

Ouvrir le JSON brut

−69.1%réduction médiane de tokens (mesurée)

Médiane sur 7 codebases réelles (1,157 fichiers).

updated 2 days agobf25a0c

Source : public/assets/benchmarks.json

−69.1%tokens en moins par tâche (médiane)

≈3.2× smallerfocused capsule vs reading the full files

25 msrequête de contexte médiane

84central-symbol tasks measured across 7 repos

Méthodologie

Comment la réduction est mesurée.

For each repo, the top-12 symbols by PageRank are used as tasks. Baseline = full source of the files containing the search hits (what an agent opens). Capsule = AIndex focused context at a 32k-token budget. Reduction = 1 − capsule/baseline (median).

Projet	Fichiers	Tokens économisés	Requête	Garde-fou qualité
AiIndexer	355	−77.5%	121 ms	ok
matchr	258	−75.6%	169 ms	ok
monsite	193	−70.1%	25 ms	ok
vouch	150	−59.2%	22 ms	ok
vps-dashboard	94	−69.1%	66 ms	ok
tally	55	−54.2%	13 ms	ok
loop	52	−58.1%	14 ms	ok
Total / médiane	1,157	−69.1%	25 ms	ok

AiIndexerFichiers355Tokens économisés−77.5%Requête121 ms

matchrFichiers258Tokens économisés−75.6%Requête169 ms

monsiteFichiers193Tokens économisés−70.1%Requête25 ms

vouchFichiers150Tokens économisés−59.2%Requête22 ms

vps-dashboardFichiers94Tokens économisés−69.1%Requête66 ms

tallyFichiers55Tokens économisés−54.2%Requête13 ms

loopFichiers52Tokens économisés−58.1%Requête14 ms

Réduction de tokens = capsule de contexte ciblée d'AIndex vs lecture intégrale des fichiers contenant les résultats, mesurée sur les symboles les plus centraux de chaque dépôt à un budget de 32k. Le proof vérifie aussi la qualité de compression : top hit conservé, couverture des meilleurs symboles et couverture des fichiers candidats.

Estimateur de ROI

Estimez l'économie — à partir de la médiane en direct, jamais d'un chiffre en dur.

Seule la part de lecture de contexte de chaque tâche reçoit la réduction mesurée. Toutes les hypothèses sont affichées ; le résultat est une estimation, pas une facture.

DEMO

DevelopersAgent tasks / dev / dayPrice per 1M input tokens ($)

$63estimated saved / month

$752estimated saved / year

20,895,840context tokens avoided / month

Assumptions (all editable in the math, none hidden)

Measured median context reduction: 69.1% — read dynamically from public/assets/benchmarks.json, applied only to the context-reading tokens.
~30,000 input tokens per agent task, 60% of which is code-context reading.
21 working days per month.

This is an estimate, not a measured invoice saving. Only the context-reduction percentage is benchmarked; token-per-task, context share and pricing are your inputs/assumptions. Real savings depend on your repos, agents and prompt mix.

Limite honnête

Benchmark agent end-to-end.

Il mesure une tâche d'agent complète (pass@1) entre AIndex / vexp / baseline — distinct du benchmark de réduction de tokens ci-dessus.

En cours de mesure

Aucun chiffre n'est affiché ici pour l'instant. Tant qu'une vraie campagne n'est pas publiée, ce bloc reste vide par principe plutôt que d'inventer une valeur.

La méthodologie et le statut vivent dans web/public/assets/agent-benchmark.json · status=template

Échantillon public de pilote

Un rapport de pilote anonymisé que vous pouvez inspecter.

Un pilote en dogfood, anonymisé et métadonnée-only — aucun code source brut. La même forme qu'un acheteur reçoit, sans les données client.

SAMPLE

Provenance de l'échantillonsample · dogfood, anonymized

updated 2 days agobf25a0c

Dépôts analysés3anonymisé

Réduction de contexte moyenne69.1%chiffre d'échantillon

No rawenforcedsource exclue

Ouvrir le JSON d'échantillon Export PDF

Tâches d'exempleDemo preview

Fix authentication redirect loop

8 file/symbol pivots, 3 tests, 2 risk zones — capsule replaced reading the full auth module.

Add idempotency to the billing webhook

6 pivots across handler + store, 4 tests, 1 risk zone.

Risques détectés

medium - Untested webhook retry path - add an idempotency test before rollout
low - Mixed logging conventions across services - normalize during onboarding

Déploiement recommandé

Roll out to the two highest-traffic repos first; both clear the quality gate.
Wire the MCP server into the default agent profile to close the adoption gap.
Publish this JSON/PDF bundle as the procurement proof artifact; keep raw source out by default.

Report provenanceDemo preview

Shapepreview-pilot-reportweb normalized

Hashnot observedraw hidden

No rawenforcedsource excluded

Accepte la sortie CLI native de aindex pilot --json.

Bundle d'export acheteurJSON + PDF

Artifact

`pilot-report.pdf`

Artifact

`pilot-report.json`

Artifact

`roi-summary.json`

Artifact

`adoption-checks.json`

Artifact

`governance-audit.json`

Artifact

`sourcegraph-parity-checklist.json`

Déploiement recommandé

Start with Pro for local Autopilot and Context Packs.
Move to Team when PR Packs, Team Shared Index and ROI Dashboard are used by reviewers.
Use Enterprise if RBAC, audit exports, policy firewall or air-gapped operation are required.

Runtime feature proof

Feature discovery is backed by metadata-only runtime evidence.

Public pages default to preview data. When a local SQLite, JSON or cloud summary is explicitly allowed, the same registry shows which workflows have been observed without exposing raw source.

0/26observed workflowsDemo preview, not runtime observed

0evidence item(s)no raw source included

26registered workflowsFeature Registry feeds web, CLI, MCP, dashboard, setup and VSCode.

4status fieldsEnabled, available, observed and buyer grade stay separate on public pages.

Feature visibility checklist

Proof stays attached to the full master plan.

Token reduction is the engine signal. The proof page also keeps Autopilot, Context Packs, cache-aware context, adoption, waste, governance, Team Index, PR packs, Batch Autopilot, Insights and Sourcegraph parity visible.

Agent Cost AutopilotBeta

Agent Cost Autopilot

Budgeted plan before the agent starts reading.

Surface: CLI, MCP context, dashboard, pilot report
Buyer-safe status: Beta · Pilot-grade
Runtime: preview only

Context PacksBeta

Context Packs

Graph-ranked task packs with hashes, cache classes and exact anchors.

Surface: CLI pack, MCP links, PR Pack, dashboard timeline
Buyer-safe status: Beta · Pilot-grade
Runtime: preview only

Cache-Aware ContextFoundation

Cache-Aware Context

Stable and volatile context are separated so cache behavior is visible.

Surface: Autopilot, benchmarks, ROI dashboard, pilot report
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Proof / ROI DashboardBeta

Proof / ROI Dashboard

Tokens avoided, cost saved, time saved and predicted vs observed proof.

Surface: Dashboard, Pilot PDF, pricing, procurement export
Buyer-safe status: Beta · Pilot-grade
Runtime: preview only

Loop KillerFoundation

Agent Loop Killer

Repeated reads, ignored packs and loop recovery actions are surfaced.

Surface: CLI waste, dashboard waste table, MCP status, monitors
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Adoption DetectorFoundation

Agent Adoption Detector

Configured agents are checked against recent MCP usage.

Surface: Setup success, dashboard alerts, IDE sidebar, pilot report
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Deep SearchFoundation

Deep Search

Budgeted multi-step research returns citations and a stop condition.

Surface: CLI deep-search, MCP context, saved queries, dashboard evidence
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Sourcegraph parityFoundation

Sourcegraph Parity Layer

Parity language is explicit while Sourcegraph-class claims stay gated.

Surface: Compare Sourcegraph, pricing, pilot report, benchmarks
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Team Shared IndexFoundation

Team Shared Index

One commit-addressed team index with freshness and local fallback.

Surface: Team index status, dashboard, CI, MCP status
Buyer-safe status: Foundation · Foundation
Runtime: preview only

PR Context PackFoundation

PR Context Pack

Changed files, blast radius, tests, owners and risk gate are packaged.

Surface: CI artifact, dashboard PR packs, GitHub/GitLab comment
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Governed ContextFoundation

Governed Context

RBAC, redaction, denied paths and audit exports stay visible.

Surface: Governance audit, Enterprise policy, monitors, pilot report
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Pilot ReportFoundation

Pilot Report

Buyer-ready report with baseline, examples, risks and rollout tier.

Surface: CLI pilot, PDF export, dashboard, compare pages, pricing
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Batch AutopilotFoundation

Batch Autopilot

Dry-run migration plan with rollout gate before PR creation.

Surface: CLI batch-autopilot, dashboard batch runs, owners, PR metadata
Buyer-safe status: Foundation · Foundation
Runtime: preview only

InsightsFoundation

Insights & Monitoring

Saved query snapshots become trends instead of one-off searches.

Surface: Dashboard insights, saved queries, pilot report
Buyer-safe status: Foundation · Foundation
Runtime: preview only

MonitoringFoundation

Insights & Monitoring

Pattern alerts and threshold checks are explicit product surfaces.

Surface: Monitors, CI output, webhooks, governance alerts
Buyer-safe status: Foundation · Foundation
Runtime: preview only

OwnershipFoundation

Ownership & Blast Radius

Likely owners and affected tests are attached to risky changes.

Surface: PR Pack, Batch Autopilot, dashboard, IDE current diff
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Feature DiscoveryFoundation

VS Code, CLI and MCP Surfaces

No workflow ships as a hidden command or undocumented MCP action.

Surface: Feature registry, setup success, dashboard, docs, pricing
Buyer-safe status: Foundation · Foundation
Runtime: preview only

VSCodeFoundation

VS Code, CLI and MCP Surfaces

IDE actions mirror the same registry used by CLI, MCP and web.

Surface: VS Code sidebar and command palette
Buyer-safe status: Foundation · Foundation
Runtime: preview only

CLIFoundation

VS Code, CLI and MCP Surfaces

Commands expose the same feature names, statuses and JSON contracts.

Surface: aindex features, status, context, pack and workflow commands
Buyer-safe status: Foundation · Foundation
Runtime: preview only

MCPFoundation

VS Code, CLI and MCP Surfaces

Five tools route advanced actions without schema bloat.

Surface: context, search, navigate, status and remember
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Advanced Code NavigationFoundation

Advanced Code Navigation

architecture, dead-code, smells, audit, duplicates, implementations, type-hierarchy, map and rules are one router.

Surface: MCP navigate, CLI, docs, dashboard, architecture overview
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Cross-Session MemoryFoundation

Cross-Session Memory

Observations and run outcomes persist and are recalled across sessions, stored locally.

Surface: MCP remember, CLI, docs, dashboard, agent sessions
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Batch Search ToolsFoundation

Batch Search Tools

search action=batch groups related queries under one budget to cut round-trips.

Surface: MCP search, CLI, docs, dashboard, saved queries
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Self-Hosted & Air-GappedFoundation

Self-Hosted & Air-Gapped

Local SQLite index, no-network mode and metadata-only exports prove zero egress.

Surface: Security, Enterprise policy, CLI doctor, docs, dashboard
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Code Graph ExportFoundation

Code Graph Export

SCIP/LSIF interop and native full-fidelity .aig export/import keep the graph portable.

Surface: CLI scip, CLI graph, docs, dashboard, interop
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Auto-WikiFoundation

Auto-Wiki

A deterministic, metadata-only code wiki is rendered from the live graph with signatures and flows but no raw source.

Surface: CLI wiki, MCP navigate, docs, dashboard, architecture overview
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Change CouplingFoundation

Change Coupling

Git co-change pairs surface coupled neighbors a static dependency graph misses.

Surface: CLI context, MCP context, docs, dashboard, PR Pack
Buyer-safe status: Foundation · Foundation
Runtime: preview only

One-Call Context PipelineFoundation

One-Call Context Pipeline

One context call returns capsule, impact, memory recall and change-coupling together.

Surface: CLI context, MCP context, docs, dashboard, agent sessions
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Enterprise Access ControlFoundation

Enterprise Access Control

RBAC, SSO and SCIM are Enterprise entitlements enforced through governance policy.

Surface: Enterprise policy, governance audit, security, pricing, docs
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Codebase AtlasFoundation

Codebase Atlas

A deterministic, metadata-only system model, drift report and ranked risk board for any AI-built repo.

Surface: CLI understand, MCP status, Studio Atlas home, dogfood page, docs
Buyer-safe status: Foundation · Foundation
Runtime: preview only

Buyer surfaces

Billing, security, benchmarks and pricing have explicit guardrails.

These surfaces support the feature story without turning foundations into shipped enterprise claims.

Feature registryLive

Every master-plan workflow is named, linked, tiered and tied to runtime feature proof.

The registry is generated from the same source used by pricing, dashboard, setup and MCP status.

Open surface

ROI DashboardFoundation

Autopilot runs, PR packs, team index freshness, adoption, waste, insights, monitors and governance are visible as one command center.

Demo preview and observed data stay clearly separated, with metadata-only runtime evidence.

Open surface

PricingLive

Plan tiers map to named workflows with runtime observed/available kept separate.

No plan page should imply Sourcegraph-class parity for foundation workflows.

Open surface

BillingFoundation

Checkout and portal paths are visible, with readiness gates and support fallback.

Billing is presented as guarded until live-sales readiness confirms checkout.

Open surface

SecurityFoundation

Local-first source posture, redaction, audit and Governed Context are connected.

Security copy stays metadata-only and avoids claiming hosted source indexing.

Open surface

BenchmarksLive

Public benchmark data is tied back to cache-aware context and pilot proof.

Benchmark pages distinguish measured local runs from positioning claims.

Open surface

Compare SourcegraphFoundation

Search, graph, Deep Search, Batch, Insights, Monitoring, Ownership, governance and ROI are mapped to registry-backed workflows.

The page names parity expectations while keeping Sourcegraph-class claims gated by availability and runtime evidence.

Open surface

Compare Cursor/CodyFoundation

AiIndexer is presented as the neutral MCP context layer under Cursor, Cody and other assistants.

The comparison avoids replacing IDE assistants; it focuses on shared packs, adoption checks, governance and ROI proof.

Open surface

Pilot reportFoundation

Autopilot, packs, adoption, governed context, team index, Sourcegraph parity and pricing recommendations become one buyer artifact.

Observed pilot inputs stay metadata-only, with demo preview clearly separated from real report data.

Open surface

Workflows de preuve

La preuve devient un argument commercial quand le run d'agent est mesuré de bout en bout.

Ces fonctionnalités transforment la réduction de tokens en coût évité, preuve d'adoption, stabilité du cache et export de pilote prêt pour l'acheteur.

ProFoundation

Cache-Aware Context

Order context so prompt caching can actually hit.

The buyer sees that AiIndexer reduces tokens and protects prompt-cache hit rate at the same time.

aindex cache "refactor auth callbacks" --jsonSee feature

TeamBeta

Proof / ROI Dashboard

Show tokens avoided, cost saved, time saved and wasted reads.

Management gets numbers it can forward: tokens avoided, cost saved, time saved and loops killed.

aindex roi --jsonSee feature

ProFoundation

Agent Loop Killer

Detect repeated reads, ignored packs and failing-test loops.

The dashboard catches the exact behavior teams complain about: repeated reads and tests without progress.

aindex waste --ingest agent-session.jsonl --source codex --jsonSee feature

ProFoundation

Agent Adoption Detector

Verify Codex, Claude Code, Cursor, Windsurf, VS Code, Zed, Copilot and Amp really call MCP.

Install success stops being a guess because every agent is checked against real MCP usage.

aindex adoption --repair --jsonSee feature

TeamFoundation

Pilot Report

Generate a 30-minute report on the customer's real repos.

Sales can prove ROI on the customer's code before procurement turns the conversation into belief.

aindex pilot --repos repo1 repo2 repo3 --pdf --json --json-output pilot-report.jsonSee feature

TeamFoundation

Sourcegraph Parity Layer

Cover the indispensable code intelligence workflows, then win on agent ROI.

The buyer recognizes the enterprise checklist, then sees AiIndexer win on agent cost proof.

aindex features --jsonSee feature

Au-delà du contexte

AIndex fait bien plus que récupérer du contexte.

Même cœur graphe-contexte, plus une couche complète d'audit & d'analytics que vexp ne propose pas. D'après les fonctionnalités annoncées par vexp (vexp.dev).

Capacité	AIndex	vexp
Capsule de contexte graphe-aware	✓	✓
Analyse d'impact / blast-radius	✓	✓
Mémoire persistante entre sessions	✓	✓
Traçage du chemin d'exécution	✓	✓
Détection de code mort	✓	—
Détection de code dupliqué	✓	—
Code smells & audits	✓	—
Trouve ton code le plus critique	✓	—
Analytics d'économie de tokens	✓	—
Licence offline / air-gapped	✓	—

Capsule de contexte graphe-awareAIndex✓vexp✓

Analyse d'impact / blast-radiusAIndex✓vexp✓

Mémoire persistante entre sessionsAIndex✓vexp✓

Traçage du chemin d'exécutionAIndex✓vexp✓

Détection de code mortAIndex✓vexp—

Détection de code dupliquéAIndex✓vexp—

Code smells & auditsAIndex✓vexp—

Trouve ton code le plus critiqueAIndex✓vexp—

Analytics d'économie de tokensAIndex✓vexp—

Licence offline / air-gappedAIndex✓vexp—

Chiffres vexp : vexp.dev et le dépôt vexp-swe-bench (ils annoncent aussi 73% de pass@1 sur SWE-bench, et plus de langages/agents). Les méthodologies diffèrent — un face-à-face sur le même banc est prévu. · vexp.dev

Comparatif

Coller du code dans un chat n'est pas une stratégie de contexte.

AIndex n'est pas un énième chat. C'est la couche de contexte locale qui fournit à votre agent existant exactement le code connecté dont chaque tâche a besoin.

Comment vous donnez le contexte aujourd'hui	Avantage	Limite	Avec AIndex
Copier-coller à la main	Rapide pour une seule ligne	Rate tout ce à quoi le code est lié	Renvoie automatiquement le code connecté.
Recherche par mot-clé / texte	Trouve des chaînes	Ignore comment le code est relié	Classe par le graphe de dépendances, pas seulement par le texte.
Index de code dans le cloud	Centralisé	Envoie votre code source sur un serveur	Reste 100 % local par défaut.
Envoyer des fichiers entiers	Facile	Sature la fenêtre de contexte (20–50k tokens)	Envoie une capsule à niveaux — généralement 4–8k tokens.
RAG par embeddings (recherche vectorielle)	Rappel « sémantique »	Correspondances floues ; pas de vrai graphe d'appels	Classe sur le vrai graphe de dépendances et montre l'impact — pas juste la similarité.
« Aller à la définition » de l'IDE	Sauts précis	Un symbole à la fois ; aucun contexte au niveau de la tâche	Assemble toute la capsule de la tâche en un seul appel MCP.
Autres indexeurs de code MCP	Contexte classé, local	Limitent souvent les appels quotidiens et réservent les outils de graphe (appelants, impact) au payant ; leur « sémantique » est un simple TF-IDF/mots-clés	Navigation de graphe — références, appelants, impact — incluse en Free. Appels illimités. Résultats exacts, résolus par le graphe.
Chasse manuelle aux références	Marche dans ton IDE	Ton agent IA ne peut pas le faire — il retombe sur grep (rate les appels indirects, accroche commentaires/chaînes)	Donne à l’agent le « Rechercher toutes les références », appelants/appelés et aller-à-la-définition, avec file:line exact.

Copier-coller à la mainAvantageRapide pour une seule ligneLimiteRate tout ce à quoi le code est liéAvec AIndexRenvoie automatiquement le code connecté.

Recherche par mot-clé / texteAvantageTrouve des chaînesLimiteIgnore comment le code est reliéAvec AIndexClasse par le graphe de dépendances, pas seulement par le texte.

Index de code dans le cloudAvantageCentraliséLimiteEnvoie votre code source sur un serveurAvec AIndexReste 100 % local par défaut.

Envoyer des fichiers entiersAvantageFacileLimiteSature la fenêtre de contexte (20–50k tokens)Avec AIndexEnvoie une capsule à niveaux — généralement 4–8k tokens.

RAG par embeddings (recherche vectorielle)AvantageRappel « sémantique »LimiteCorrespondances floues ; pas de vrai graphe d'appelsAvec AIndexClasse sur le vrai graphe de dépendances et montre l'impact — pas juste la similarité.

« Aller à la définition » de l'IDEAvantageSauts précisLimiteUn symbole à la fois ; aucun contexte au niveau de la tâcheAvec AIndexAssemble toute la capsule de la tâche en un seul appel MCP.

Autres indexeurs de code MCPAvantageContexte classé, localLimiteLimitent souvent les appels quotidiens et réservent les outils de graphe (appelants, impact) au payant ; leur « sémantique » est un simple TF-IDF/mots-clésAvec AIndexNavigation de graphe — références, appelants, impact — incluse en Free. Appels illimités. Résultats exacts, résolus par le graphe.

Chasse manuelle aux référencesAvantageMarche dans ton IDELimiteTon agent IA ne peut pas le faire — il retombe sur grep (rate les appels indirects, accroche commentaires/chaînes)Avec AIndexDonne à l’agent le « Rechercher toutes les références », appelants/appelés et aller-à-la-définition, avec file:line exact.

Honnête par défaut

Chaque chiffre renvoie à une commande.

Chaque benchmark est généré par le binaire lors d'un run local — taille du dépôt, symboles, arêtes et vitesse de ré-indexation, sans rien retoucher à la main.

aindex autopilot "fix billing login bug" --json
aindex pilot --repos repo1 repo2 repo3 --pdf --json
aindex roi --json
aindex features --json
ai-indexer-mcp proof --repo ~/Documents/matchr --json

Ouvrir le JSON brut

Dernier run

Le JSON exact, directement depuis l'outil.

Loading benchmark data…

Limites honnêtes

Ce que ces chiffres disent et ne disent pas.

Ce sont des temps d'indexation et de ré-indexation locaux sur une seule machine, pas un benchmark hébergé. Vos chiffres dépendent de la taille du dépôt, du mix de langages et de la vitesse du disque — lancez la commande sur votre propre code pour voir les vrais chiffres.

No hidden features

Feature discovery is a product system, not a page.

The same registry has to surface in setup, CLI, MCP status, dashboard, docs, pricing, release notes and IDE actions so users cannot miss the value they already installed.

Setup success

Run Autopilot, Generate PR Pack, Check Adoption, Run Pilot Report, Open ROI Dashboard

The first successful install immediately points users to money-saving workflows.

Feature registry

Name, punchline, problem, command, MCP action, plan, maturity, proof and outcomes

Marketing, docs, pricing, setup, dashboard and release notes share the same product truth.

CLI

autopilot, pack, cache, deep-search, pr-pack, team-index, pilot, adoption, waste, governance, batch, insights, monitors, ownership

Users can discover and run workflows without reading MCP schemas or source code.

MCP

context, search, navigate, status and remember route every advanced action internally

The surface stays at five tools while still exposing enterprise workflows.

Dashboard

ROI cards, run timeline, PR packs, team index, adoption alerts, waste, insights, monitors, batch runs, governance audit

The buyer sees savings and governance evidence instead of a generic search console.

IDE

Autopilot, Pack, Cache, Deep Search, PR Pack, Loop Killer, Adoption, Team Index, ROI, Governance, Pilot, Insights, Monitoring, Batch, Ownership, Blast Radius

Agents and developers see the same actions where coding work happens.

Setup success screen

After install, the product should push users into value immediately.

These are the first five actions that make the platform visible before users forget the MCP exists.

Setup action

Run Autopilot

Shows naive tokens, optimal pack, risks, tests and do-not-read paths.

aindex autopilot "fix billing login bug" --json

MCP: context action=autopilot task="fix billing login bug"

Setup action

Generate PR Pack

Turns review into files touched, blast radius, owners, tests, risks and publish-safe comment metadata.

aindex pr-pack --pr 42 --repo owner/name --provider github --publish-comment --dry-run --json

MCP: context action=pr-pack pr=42 repo=owner/name

Setup action

Check Adoption

Confirms each agent is configured and actually calling MCP.

aindex adoption --repair --json

MCP: status detail=adoption

Setup action

Run Pilot Report

Creates JSON/PDF management proof with provenance, no raw source, savings, adoption, risk and recommendations.

aindex pilot --repos web api billing --pdf --json --output aindex-pilot-report.pdf --json-output pilot-report.json

MCP: status detail=pilotPreview: AINDEX_PILOT_REPORT_PATH=./pilot-report.json AINDEX_PILOT_REPORT_PUBLIC_OBSERVED=1 npm --prefix web run dev

Setup action

Open ROI Dashboard

Displays observed savings, waste findings, team index state and governance audit.

aindex roi --json

MCP: status detail=roi

Feature discovery

Proof is the sales wrapper around the platform.

Autopilot, Context Packs, Adoption Detector, Waste Analyzer, Team Shared Index, Governed Context and ROI Dashboard all become visible in one proof page. Pilot Report included.

View all features

Autopilotpreview only

Agent Cost Autopilot

Give the agent a budgeted plan before it starts reading.

The first response tells the team how much context the agent would have burned and what it should read instead.

Naive context, required pack, savings, risk zones, files to read, files to avoid.

Pro · Beta · Pilot-grade · naive_tokens + required_pack_tokensEnabled: yes · Available: yes · Observed: no · Grade: Pilot-grade

Contextpreview only

Context Packs

Graph-aware task packs with exact symbol anchors, deps, tests, risks and cache classes.

A task becomes a reusable artifact that agents, CI and reviewers can consume without reconstructing context.

Each item carries fqn, path, line range, graph-aware reason, rank, token estimate, hash and cache class.

Pro · Beta · Pilot-grade · pack_items + fqnEnabled: yes · Available: yes · Observed: no · Grade: Pilot-grade

Contextpreview only

Cache-Aware Context

Order context so prompt caching can actually hit.

The buyer sees that AiIndexer reduces tokens and protects prompt-cache hit rate at the same time.

stable_tokens, volatile_tokens and cache_hit_estimate become first-class metrics.

Pro · Foundation · Foundation · stable_tokens + volatile_tokensEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Proofpreview only

Proof / ROI Dashboard

Show tokens avoided, cost saved, time saved and wasted reads.

Management gets numbers it can forward: tokens avoided, cost saved, time saved and loops killed.

Predicted vs observed savings, top repos, top agents, loops and export-ready summary.

Team · Beta · Pilot-grade · tokens_avoided + cost_savedEnabled: yes · Available: yes · Observed: no · Grade: Pilot-grade

Wastepreview only

Agent Loop Killer

Detect repeated reads, ignored packs and failing-test loops.

The dashboard catches the exact behavior teams complain about: repeated reads and tests without progress.

Findings include ingested agent transcripts, repeated files, ignored MCP, repeated tests and suggested next calls.

Pro · Foundation · Foundation · loop_findings + wasted_readsEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Adoptionpreview only

Agent Adoption Detector

Verify Codex, Claude Code, Cursor, Windsurf, VS Code, Zed, Copilot and Amp really call MCP.

Install success stops being a guess because every agent is checked against real MCP usage.

Reports configured yes/no, last MCP call, likely issue and idempotent repair command.

Pro · Foundation · Foundation · agent + configuredEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Teampreview only

Team Shared Index

One team index per commit instead of every dev indexing the same repo.

A CTO sees one commit-addressed index serving the team instead of every developer re-indexing locally.

Fresh/stale status, commit SHA cache, team ROI report and fallback to local if the server is unavailable.

Team · Foundation · Foundation · commit_sha + freshnessEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Reviewpreview only

PR Context Pack

Every PR gets provider-fetched changed files, blast radius, tests, owners and risks.

Reviewers get changed files, blast radius, tests, owners and sensitive risks before opening the diff.

Metadata-only provider context from gh/glab when available, CI artifact, review gate decision and optional GitHub/GitLab comment with reviewer summary.

Team · Foundation · Foundation · changed_files + provider_contextEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Governancepreview only

Governed Context

RBAC, redaction, audit logs and policy filters for agent context.

Security can prove which agent asked for which context and which paths were denied or redacted.

Audit log: agent, repo, request, pack, files, policy decision, redaction status and metadata-only export.

Enterprise · Foundation · Foundation · subject + policy_idEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Sales proofpreview only

Pilot Report

Generate a 30-minute report on the customer's real repos.

Sales can prove ROI on the customer's code before procurement turns the conversation into belief.

Files, symbols, reduction estimate, risks, adoption and baseline comparison.

Team · Foundation · Foundation · repos_scanned + symbols_mappedEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Searchpreview only

Deep Search

Iterative local research with evidence, citations and a stop condition.

Complex code questions become budgeted research with citations instead of another open-ended search loop.

Steps, evidence, citations, budget used and no-loop stop condition.

Pro · Foundation · Foundation · steps + evidenceEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Change managementpreview only

Batch Autopilot

Plan multi-repo changes with an enterprise rollout gate before opening PRs.

Teams get Sourcegraph-style migration planning without surprise PR storms or opaque rollout cost.

Repos touched, candidate patches, tests, owners, risks, PR metadata and a metadata-only rollout gate with CI conclusion.

Team · Foundation · Foundation · target_repos + candidate_changesEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Governancepreview only

Insights & Monitoring

Saved queries, trends and monitors for smells, deprecated APIs and token waste.

Saved searches become weekly signals for deprecated APIs, risky files, token waste and agent adoption.

Snapshots, trends, alerts and CI/webhook output.

Team · Foundation · Foundation · saved_query + snapshotEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Reviewpreview only

Ownership & Blast Radius

Resolve owners and show what a change can break.

Every risky change can name likely owners, downstream impact and the tests that protect it.

CODEOWNERS, git blame, path rules and team mapping feed PR packs and governance.

Team · Foundation · Foundation · owners + affected_symbolsEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Competitivepreview only

Sourcegraph Parity Layer

Cover the indispensable code intelligence workflows, then win on agent ROI.

The buyer recognizes the enterprise checklist, then sees AiIndexer win on agent cost proof.

Search, navigation, Deep Search, Batch, Insights, Monitoring, Ownership, permissions and API surfaces.

Team · Foundation · Foundation · parity_rows + wedge_rowsEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Adoptionpreview only

VS Code, CLI and MCP Surfaces

The same structured JSON powers CLI, MCP tools and IDE sidebars.

Every feature appears where agents are actually used: setup, CLI, MCP status, dashboard and IDE.

Feature registry feeds docs, CLI, MCP, dashboard, setup success and VS Code sidebar.

Free · Foundation · Foundation · feature_id + commandEnabled: yes · Available: yes · Observed: no · Grade: Foundation

Navigationpreview only

Advanced Code Navigation

Architecture overview, dead code, smells, audit, duplicates, implementations, type hierarchy, map and rules in one router.