AI & ML
almost-cli — BYOK Multi-Agent Coding, Local First
Parth Gorde Dev.to (EN Zone)
3 views
Repo: https://github.com/wellparth/almost-cli
Most coding assistants rent you a model and keep your code in the cloud. almost-cli takes a different approach: bring your own API keys, run a team of specialized agents locally, and keep everything under your control.
The idea
almost-cli is a provider agnostic CLI for multi-agent coding. Connect your own OpenAI, Gemini, DeepSeek, or NVIDIA NIM keys and orchestrate agents: Architect, Planner, Builder, Tester, Debugger against a local codebase. No hosted inference. No backend account. Your keys stay on your machine.
The core insight is simple: agent identity is separate from model identity. An agent is a configuration of system prompt, tool set, permissions, and workflow. The model is just the runtime. You can route cheap models for planning and strong models for debugging without changing behavior.
How it works
The stack is layered for safety and composability:
CLI/TUI talks to an Agent Orchestrator
Orchestrator coordinates Agent Runtime, Context Engine, and Permission Engine
Agents run with explicit tool sets: Filesystem, Shell, Git, Search, MCP
Permissions are deny by default. Every tool call is checked. MCP servers are permission gated per tool.
This makes local first multi-agent work practical. You get auditability, explicit boundaries, and the ability to swap providers per agent.
Why BYOK matters
With BYOK you own your data, your keys, and your agent definitions. There is no vendor lock in and no code leaves your machine unless you ask for it. The trade off is you own ops, which is exactly the point for teams that care about security and control.
Repo layout
apps/cli
packages/agent-core
packages/providers/*
packages/tools/*
packages/permissions
packages/context
packages/orchestrator
packages/sessions
packages/storage
packages/agent-definitions
agents/*
Try it
pnpm install
pnpm typecheck
pnpm test
pnpm build
If you build developer tools, agent security, or local first infra, I would love feedback. The repo is open and I am iterating in public.
Today was day one of building. If you find any issues, please open them. #buildinpublic
Read original: https://dev.to/wellparth/almost-cli-byok-multi-agent-coding-local-first-37nf
← Previous
We Wrote a JSON Parser from Scratch in 72 Hours to Prove We Still Could
Next →
Every Way LLM Provider Failover Breaks in Production
Related
The Difference Between an AI Agent That Works and One You Can Trust
AI & ML
0
DEV Community
The Agent Loop Nobody Talks About: Think, Act, Observe, Repeat
AI & ML
0
DEV Community
How you frame a question changes what an LLM actually argues, not just its tone
AI & ML
0
DEV Community
An LLM judge cannot be a build gate, and it is not about the cost
AI & ML
0
DEV Community
Comments0
No comments yet — be the first