Multi-agent coordination
synlynk routes coding tasks
across AI vendors and local models.
Synlynk is a neutral control plane that routes coding tasks across AI vendors and local models, then proves the result.
synlynk in action
One command. Four agents. One project.
Dispatch, monitor, and prove coding work across AI vendors and local models from a single terminal.
The multi-agent unlock
You're not limited to one agent.
You just needed a harness.
Every AI CLI is powerful in isolation. But they don't share context, don't know about each other, and don't report costs in one place. synlynk wires them up.
Context injection
Every agent gets your project memory, roadmap, and current todos — automatically, before each invocation.
Unified cost tracking
Token counts, per-agent spend, and budget limits in one dashboard. No more surprise bills.
Sentinel monitoring
Detects hallucination loops, quota hits, and consecutive failures. Alerts before they compound.
Claude alone: ~$0.015/1K tokens · Distributed across 3 agents: ~$0.006/1K tokens effective rate · ~60% cheaper effective rate when distributed across 3 agents
Getting started
Up in four commands.
synlynk is a single-file Python CLI. No build step. No dependencies. Works with whatever AI CLIs you already have installed.
STEP 01
Install
One pipx command. Installs synlynk into an isolated venv and puts the binary on your PATH.
pipx install git+https://github.com/nikhilsoman/synlynk
STEP 02
Init
Bootstrap your project — creates project-docs/, config, and AI instruction files.
synlynk init
STEP 03
Exec
Run any AI CLI through synlynk — context is injected, costs tracked, output monitored.
synlynk exec claude
STEP 04
Dispatch
Parallelize tasks across multiple agents from one command. Results flow into shared project state.
synlynk dispatch
claude agy codex
--task "ship it"
claude agy codex
--task "ship it"
Feature spotlight
Four things that make the agent network work.
Dispatch
Run a task across multiple agents in parallel. Each agent gets the same context snapshot. Results stream into a unified log with per-agent cost attribution.
synlynk dispatch claude agy codex \
--task "review PR #47" \
--story SL-12 --context-mode team
Context Injection
Before every exec or dispatch, synlynk builds a context snapshot from your project docs — memory, roadmap, todos, and teammate devlogs — and injects it into the agent's prompt.
synlynk context --mode team
# → memory + roadmap + devlog/nikhil
# → 2,341 tokens injected
Sentinel Monitoring
Real-time pattern detection on agent output. Flags FLATLINE (3 consecutive failures), SUCCESS_LOOP (identical outputs), QUOTA_EXHAUSTED, and other anomalies. Alerts go to sentinel.md.
synlynk sentinel --watch
⚠ SUCCESS_LOOP: job-d1e4
3× identical output detected
Agent Profiles
Each agent has a profile describing its strengths, rate limits, and image-gen capabilities. Profiles inform dispatch routing — Imagen tasks go to Agy, DALL-E to Codex, Aurora to Grok.
synlynk profile set agy \
--role image-gen \
--model imagen-3
The coordination layer
Synlynk is a neutral control plane for coding tasks.
Dispatch, relay, monitor, and account for every agent in your org — from a single shared project state.
4
agents
0
deps
MIT
license
Documentation
Everything you need to get started.
⏱️
Quick Start Guide
Get synlynk up and running in under five minutes with our step-by-step setup guide and example project.
Read Guide →💻
Command Reference
Detailed syntax and examples for every command including exec, join, dispatch, status, and doctor.
View Reference →📖
The Manual
A deep dive into state.db schemas, agent role definitions, budget enforcement, and relay network architecture.
Explore Manual →Early access
Get early access to synlynk.com
Account management, team workspaces, and hosted dispatch — coming soon.
Thanks — you are on the list. (synlynk.com account portal coming soon)