Skip to content
Vectalon
Get started

Vectalon

Adaptive AI harness for developers

◈Local-first◎Project-aware◆Deterministic▣Control plane
See it runCompare plans
▣vectalon main
[ v0.15.0 ]

Point it at your repo. Vectalon understands, reviews, and validates your code.

Local-first MCP harness with deterministic commands. Model providers and support uploads are optional.

intel feedlive
guardrails
  • Pressable over TouchableOpacity✓ pass
  • no leaked renders✓ pass
  • New Architecture hazards✓ pass
  • bundle budget deltas◈ watch
hallucination-verified · in editor + review
healing logvectalon feature
  • ◆ [8/14] Implementation ▸ yarn test ✓ (1.8s)
  • ✔ Compile-checked: 0 errors after 3 healing passes
  • ✦ 14/14 phases. index.md: 5 lessons distilled
  • ◆ product-manifest.json: contract v1.0.0 pinned
  • ✔ 44 agents · 64 tools · 43 scenarios verified

vectalon@main:~$ npx --yes --package=@vectalon-dev/rn@latest vectalon init

See it run
one harness, four platforms
react-nativeliveiossoonandroidsoonfluttersoon
one mcp server, every client
claude codecursorcopilotcodexgeminizedwindsurfopencode

One loop, three commands

The whole workflow (context, generation, verification) is a CLI loop. No IDE plugin, no dashboard, no state to sync.

vectalon - session
$ vectalon init
[ step 01 / 03 ]

Scan & seed

Point it at your repo. Vectalon scans structure, dependencies, and conventions, then builds the project knowledge graph.

$ vectalon serve
[ step 02 / 03 ]

The agent loop

A local MCP-aware model works your codebase with 64 project-aware tools, re-seeding ecosystem intel every hour.

$ vectalon feature "…"
[ step 03 / 03 ]

Generate & heal

Describe a task. You get compile-checked code, tests, and a code review. Fixes are typechecked and reverted if they don't reduce errors.

Watch it run: 85 seconds, no cuts

The real CLI on a real 19-screen Expo app: init, arch, sec, feature, bench.

demo/full-demo.mp4 — v0.2.0
94%
guardrail pass rate
44
registered deterministic commands
64
project-aware tools
43
benchmark scenarios
Full leaderboard: benchmark scenarios, measured against human references →

What it does for you

One harness for the whole loop (context, generation, verification, upgrade) with the model and knowledge base maintained by Vectalon, not you.

MCP-native agent

An experimental, opt-in local-model agent can use 64 registered project-aware tools. The catalog also tracks 44 deterministic commands; lifecycle, evidence, and availability vary by command.

Self-maintaining knowledge

Beta onboarding can inspect local project context. Self-maintaining knowledge and hourly refresh workflows remain experimental and require explicit opt-in.

Always-current model

Experimental web-intel inputs can augment a configured model. They depend on network sources and are not evidence of future-release knowledge.

Upgrade Copilot

Experimental upgrade tools can inspect diffs and propose bounded changes. Verification must be reviewed in the target project.

Guardrails on save

Platform best-practices rules (Pressable, no leaked renders, New Architecture hazards) run in your editor and in code review. Hallucination-verified findings.

Compile-checked healing

Experimental repair workflows can run configured type checks and bounded retries. Results still require review; this is not a guarantee that every generated fix compiles.

Impact regression coverage

Experimental analysis can map changed files to affected screens and propose regression coverage. Availability and evidence come from the released catalog.

Your platform next

iOS, Android, and Flutter harnesses are in development. Join the waitlist and we'll email the moment a beta opens.

Join the waitlist

44 registered deterministic commands. Experimental preview

The broad analysis fleet is experimental today: opt-in is required, results need review, and offline or network behavior depends on the command and configuration.

Phase 8

Autonomous engineering

Review the diff, audit the architecture, scan for secrets, diagnose a broken build or test run, propose safe refactors. For provably-safe fixes, apply them.

reviewarchsecbuild-fixtest-repairrefactordepsa11yrelease-readybug-fix
Phase 9

Release engineering

Classify crashes, score the architecture, police CI/CD workflows, check store surfaces, gather SOC 2 evidence, and aggregate everything into one executive dashboard.

crasharch-scorecicdapp-storesoc2tokensteam-statspermsdashboard
Phase 10

Enterprise intelligence

Sync design to code, rank crash classes, audit instrumentation, verify org evidence, predict release risk, and validate training data before you spend GPU time.

figmasentryobservabilitygovernanceauditreposrelease-predictplay-storedatasetlora
Full agent catalog: every command, every verdict, every report →

The model is never stale

Vectalon refreshes its own web intel every hour under serve and inlines the headlines into every model system prompt. If the ecosystem ships it, the model knows it.

See what shipped in v0.22.4 →
live intel sources26 live

Questions, answered

The short version of everything.

What do I need to run Vectalon?

Node.js ≥ 20.12 on macOS or Linux. npx --yes --package=@vectalon-dev/rn@latest -- vectalon init scans your repo and builds the knowledge graph; after installing @vectalon-dev/rn in your project, npx vectalon serve starts the MCP server your agent connects to.

Does it need a model?

Model-driven workflows require a configured provider. The catalog tracks 44 deterministic command registrations, but broad analysis and report commands are experimental. Some commands make zero model calls; network and credential requirements depend on the command and configuration.

Do the deterministic agents go stale?

No. Every run reads the project as it is right now. Source files, git history, CI config, and telemetry exports are scanned at run time. The ecosystem knowledge auto-refreshes hourly under vectalon serve.

Which platforms are supported?

React Native is live at v0.22.4. iOS, Android, and Python harnesses are in development. Join the waitlist and we'll email the moment a beta opens.

How is it different from other AI coding tools?

It isn't an AI coding assistant. It is an engineering control plane. Give Vectalon a React Native repository and it continuously understands, reviews, diagnoses, upgrades, and validates it. Deterministic and compile-checked.

What does it cost?

Free includes qualified capabilities listed by the released catalog, with no card. Experimental commands require opt-in and are not a purchased promise. Individual is $19/dev/mo and Team is $49/dev/mo.

vectalon - install

Try it. Free, no signup.

The free tier is genuinely useful: init, serve, feature, doctor. Premium commands offer a 14-day trial with one GitHub login.

$ npx --yes --package=@vectalon-dev/rn@latest -- vectalon init

Get VectalonCompare plans