v0.2-alpha · MIT · Open Source

VESPERA orchestrates AI tools as replaceable workers.

A local, open-source runtime for context, routing, and workflow control across Claude, Grok, and local models.

Try the CLI → View on GitHub
v0.2-alpha Python MIT License Windows / Mac / Linux
vespera — orchestrator
$ vespera run "audit auth module" \
  --tool claude --tool review
 
→ claude      analyzing ......... ✔ (2.1s)
→ review      verifying ......... ✔ (0.8s)
 
✔  Pipeline complete  [3/3]  2.9s
$
In practice
Cursor wrote the plan. Claude reviews it. A local agent runs tests. VESPERA keeps all three working
from the same project memory.
The Problem

Your AI stack is
working against you

The more AI tools you adopt, the more time you spend managing them — not building. Three patterns are silently killing your productivity.

01

Context fragmentation

Each AI tool maintains isolated context. You repeat yourself every session, watch Claude forget what Cursor knows, and lose the thread between tools. Your intent evaporates at every boundary.

02

Vendor lock-in by design

Every tool wants to be your primary interface — proprietary plugins, closed APIs, pricing cliffs that trap your prompts inside their ecosystem. Switching costs compound until you stop switching.

03

No composable layer

You have a dozen hammers with no way to connect them. Chaining tools means manual copy-paste or brittle scripts. There's no standard protocol for "tool A does X, then hands off to tool B."

The Solution

One protocol to
orchestrate them all

VESPERA is a lightweight orchestration runtime that treats AI tools as stateless workers — composable, observable, and replaceable on demand.

FEATURE · 01

Unified Context Bus

A shared, persistent context store all your AI tools read from and write to. One source of truth for your intent, project state, and working memory — no more cross-tool amnesia.

ctx.read() / ctx.write()
FEATURE · 02

Tool Composability Engine

Define pipelines that chain tools with typed inputs and outputs. Claude for reasoning, a local model for codegen, a custom agent for verification — all in one declarative config.

pipeline.yaml
FEATURE · 03

Zero-Lock Runtime

Open protocol, open source. Swap models, providers, or entire tool layers without rewriting your workflows. Your prompts and logic remain yours — not locked inside any vendor's SDK.

--provider any
FEATURE · 04

Observable Pipelines

Every tool call is logged, traceable, and replayable. Debug where context was lost, measure latency per tool, and compare runs side by side. Full audit trail out of the box.

vespera trace --run <id>
Architecture

You are the OS

VESPERA sits between you and every AI tool — a thin, transparent orchestration layer you always control.

YOU VESPERA CORE Context Bus · Pipeline Engine · Router Shared Memory Claude API Code Agents Local Models Custom Tools plug in anything — swap out anything
Read the Docs →
Quick Start

Up and running
in three steps

No account. No cloud setup. Runs locally from the first command.

1

Install VESPERA

bash
$ npm install -g vespera
# or with bun
$ bun add -g vespera
2

Initialise your workspace

bash
$ vespera init

✔  Created  .vespera/config.yaml
✔  Created  .vespera/pipelines/default.yaml
✔  Context bus ready on localhost:7432
3

Run your first pipeline

bash
$ vespera run "refactor auth module" \
  --tool claude \
  --tool codereview \
  --context ./src/auth

→  claude        reasoning ........ ✔
→  codereview    verifying ........ ✔
✔  Pipeline complete  (4.2s)
Check system status
bash
$ python core/vespera.py status
┌─────────────────┬──────────────────┐
│ Approval Queue  │ ✓ Clean          │
│ Asset Inbox     │ 2 pending        │
│ Today's Intake  │ 3 tasks          │
│ Last Backup     │ 2026-05-27 08:00 │
└─────────────────┴──────────────────┘
1

Install VESPERA

bash
$ npm install -g vespera
# or: curl -fsSL https://vespera.sh | sh
2

Initialise your workspace

bash
$ vespera init
✔  Context bus ready on localhost:7432
3

Run your first pipeline

bash
$ vespera run "refactor auth module" \
  --tool claude --context ./src
✔  Pipeline complete  (4.2s)
Check system status
bash
$ python core/vespera.py status
┌─────────────────┬──────────────────┐
│ Approval Queue  │ ✓ Clean          │
│ Asset Inbox     │ 2 pending        │
│ Today's Intake  │ 3 tasks          │
│ Last Backup     │ 2026-05-27 08:00 │
└─────────────────┴──────────────────┘
1

Install VESPERA

powershell
> npm install -g vespera
# or with bun
> bun add -g vespera
2

Initialise your workspace

powershell
> vespera init
✔  Context bus ready on localhost:7432
3

Run your first pipeline

powershell
> vespera run "refactor auth module" `
  --tool claude `
  --context ./src
✔  Pipeline complete  (4.2s)
Check system status
powershell
> py core/vespera.py status
┌─────────────────┬──────────────────┐
│ Approval Queue  │ ✓ Clean          │
│ Asset Inbox     │ 2 pending        │
│ Today's Intake  │ 3 tasks          │
│ Last Backup     │ 2026-05-27 08:00 │
└─────────────────┴──────────────────┘

AI tools are workers,
not owners.

VESPERA puts the final authority back where it belongs.