Vertebra — AI codebase context engine for developers

Vertebra / Anatomical Context Map

Scroll-controlled system inspection

Tracking
00 / VERTEBRASYSTEM ENTRY
Vertebra.

The architecture between your code and its intelligence.

Repeated context costUp to 90% less
Directory scan~12,000 files/s
Idle footprint< 15 MB
Time to first token< 40 ms
Local state100% on-device

Context is infrastructure

Your codebase should not be a recurring expense.

Cache active

Most coding assistants resend the same repository context with every prompt. Vertebra registers large contexts once, recalls them by signature, and bills only the work that changed.

Six systems / One command surface

Built around the work.
Not around the cloud.

SYS_01

Semantic Context Cache

Vertebra fingerprints the codebase, persona, and model into one SHA-256 signature. Matching context is recalled instead of uploaded and billed again.

SYS_02

Local by Architecture

Workspace indices, histories, and telemetry remain inside a WAL-mode SQLite database. Repository sync is never part of the system.

SYS_03

Zero-Lag File Watcher

A multi-threaded Rust watcher detects writes and deletions within milliseconds, keeping the active context synchronized with the work in front of you.

SYS_04

Context Modules

Map the exact frontend, backend, documentation, or system files a task requires. Include global standards once and activate them anywhere.

SYS_05

Native Multimodal Input

Paste screenshots, wireframes, and visual references directly beside your instruction. Image context travels with the selected codebase modules.

SYS_06

Editorial Terminal

A purpose-built Markdown lexer renders long technical answers, line-numbered code, tables, and images without a heavy parsing runtime.

Context pipeline / Deterministic

One spine.
Every signal.

Vertebra converts selected project structure into a clean reasoning substrate—then chooses the cheapest valid way to transmit it.

Five deterministic operations
One continuous context path

L1

Collect

Read only the modules and global context enabled for the current task.

L2

Serialize

Wrap topology and source files in an explicit XML reasoning boundary.

L3

Fingerprint

Hash model, persona, and payload into one deterministic context signature.

L4

Cache

Register large contexts once; reuse the active six-hour semantic lease.

L5

Transmit

Stream the response while recording precise usage metadata locally.

Structured context / XML sandbox

Boundaries the model can understand.

Files, topology, global standards, and the active persona are separated explicitly—so source code stays source code and the instruction stays the instruction.

context_payload.xml cache_hit
01 <workspace_topology root="VERTEBRA">
02  src/tauri/database.rs
03  src/tauri/fs_agent.rs
04  src/components/MarkdownBlock.tsx
05 </workspace_topology>
06 
07 <global_context_file name="ULIX_DESIGN_SYSTEM.md">
08  Editorial terminal constraints loaded.
09 </global_context_file>
10 
11 <cache_state
12  signature="sha256:5f2...c91"
13  lease_remaining="273m"
14  cached_tokens="1,222,867"
15 />
16 

Native desktop / React 19 + Tauri v2

Get back in the loop.

No browser tabs. No repository upload ritual. Vertebra keeps the project ledger, module map, personas, model controls, telemetry, and technical output inside one exacting native interface.

Vertebra / 01 / Workspace
Local rendering
The entire codebase, under control. — Vertebra application interface
01 / Workspace

The entire codebase, under control.

Map modules, switch personas, select the model, and transmit complete project context without leaving the native workspace.

Zero-trust credential boundary

Your keys belong to the
operating system.

Native keychain

macOS Keychain, Windows Credential Manager, and Linux Secret Service.

No flat-file secrets

API keys are requested on demand and never written into application storage.

Workspace isolation

Histories, mappings, and state are partitioned by local workspace identity.

Self-healing cache state

Dead remote cache handles are purged automatically before inline recovery.

Compiled specification / v0.1.0

Lightweight by construction.

Native coreRust / Tauri v2
State engineSQLite / WAL
Async runtimeTokio
File eventsNotify v8
InterfaceReact 19
StylingTailwind v4
Cache floor4,096 tokens
Cache lease6 hours

_

Intelligent Software for an intelligent future.