Spryv Agent

The autonomous coding agent that works with any model.

Give Spryv a task. It explores your codebase, diagnoses the root cause, writes the fix, runs the tests, and reports back — without you steering every step. The full autonomy of Claude Code, with the freedom to use any model you choose.

Why Spryv

Model freedom — the thing the others don't give you

Claude Code works only with Claude. Cursor locks you to their subscription. Spryv works with any OpenAI or Anthropic-compatible API — Claude, Gemini, DeepSeek, Kimi K2, any model on OpenRouter, Vertex AI, or a private self-hosted endpoint. Swap providers with a single flag. No lock-in, no walled garden, no subscription.

spryv --provider deepseek "refactor the payment module"
spryv --provider gemini "fix the race condition in the auth flow"
spryv --provider openrouter "add rate limiting to the API"

Full agent architecture — not a chat wrapper

Spryv runs a structured loop modeled on how a senior engineer actually works:

1. Explore — reads relevant files, maps call graphs, traces execution paths
2. Plan — builds an explicit plan before touching anything; reasoning is visible
3. Execute — implements changes through an architect review pass before applying each patch
4. Verify — runs tests, confirms results, reports exactly what changed and why

At every iteration it maintains a live state tree: confirmed facts, pending tasks, in-progress changes, sub-agent status. A health-check monitor detects loops and stalls and nudges the agent back on track.

Sub-agents and skills — built-in extensibility

Spawn parallel specialized sub-agents for exploration, planning, or code review. Each sub-agent gets its own context, tools, and iteration budget. Results merge back into the main session. Define custom agents and /slash-command skills in Markdown files — no code required.

Three interfaces, one engine

VS Code extension with a full chat panel, standalone Electron desktop app with a live state tree sidebar, and a CLI with phase-gated output. Same engine, same sessions, same capabilities across all three.

Install

VS Code

ext install faisalsalamah.spryv

Or install from the Marketplace / download .vsix from Releases.

CLI (Node.js 18+)

npm install -g @spryv/agent
spryv "fix the memory leak in the renderer"

Desktop App (Windows)

Download spryv-desktop-setup-v1.0.0-win.exe from Releases and run the installer.

Quick Start

spryv "add JWT auth to the API"                    # one-shot task
spryv --chat                                        # interactive mode
spryv --chat --resume <session-id>                  # resume a session
spryv --provider deepseek "fix the memory leak"    # switch provider
spryv --verbose "refactor the auth module"          # verbose output

Capabilities

| | | |---|---| | Any model | Any OpenAI or Anthropic-compatible API: Claude, Gemini, DeepSeek, Kimi K2, OpenRouter (100s of models), Vertex AI, private endpoints | | Tools | Bash execution, file read/write/edit, grep, glob, AST search, vision, sub-agent spawn, skill invoke | | Sub-agents | Parallel specialized agents; built-in: explore, planner, pr-review; fully customizable | | Skills | User-invocable slash commands; built-in: /commit, /explain; define your own in Markdown | | MCP servers | Connect any Model Context Protocol server | | Sessions | Full persistence — resume, fork, or replay any past session | | Cost tracking | Per-turn token metrics with cache hit rates and estimated cost per provider | | Interfaces | VS Code extension, Electron desktop app, CLI |

Documentation

| | | |---|---| | Installation | Setup for all platforms | | Configuration | API keys, providers, model selection | | Sub-Agents | Built-in agents and how to define custom ones | | Skills | Built-in skills and how to define custom ones |

Links

• Website: spryv.ai
• npm: @spryv/agent
• Issues: github.com/spryv/agent/issues

License

MIT