DEV.co
AI Frameworks · 2FastLabs

agent-squad

Agent Squad is an open-source framework for orchestrating multiple AI agents in conversations, routing queries intelligently to specialized agents while maintaining context. It runs on Python, TypeScript, and Swift—the latter bringing on-device orchestration to iOS and macOS.

Source: GitHub — github.com/2FastLabs/agent-squad
7.7k
GitHub stars
726
Forks
Swift
Primary language
Apache-2.0
License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

FieldValue
Repository2FastLabs/agent-squad
Owner2FastLabs
Primary languageSwift
LicenseApache-2.0 — OSI-approved
Stars7.7k
Forks726
Open issues60
Latest releasetypescript_1.1.0 (2026-07-01)
Last updated2026-07-07
Sourcehttps://github.com/2FastLabs/agent-squad

What agent-squad is

A polyglot agent orchestration framework using LLM-based or rule-based classifiers to route user input to specialized agents, with support for streaming responses, tool invocation (including MCP servers in Swift), supervisor coordination, and grounded two-LLM patterns to reduce hallucination. Integrates with Bedrock, Anthropic, OpenAI, and local models.

Quickstart

Get the agent-squad source

Clone the repository and explore it locally.

terminalbash
git clone https://github.com/2FastLabs/agent-squad.gitcd agent-squad# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Multi-domain customer support systems

Route customer queries (billing, technical, sales, returns) to specialized agents while preserving ticket context across handoffs.

On-device mobile AI assistants

Use the Swift runtime to build intelligent local-first chatbots on iOS/macOS with no cloud dependency, integrating native tools and MCP servers.

Fact-grounded information retrieval

Deploy GroundedAgent to deliver price checks, inventory status, or recommendations that must match real data exactly, not LLM inference.

Implementation considerations

  • All three runtimes (Python, TypeScript, Swift) maintain feature parity; choose based on deployment target—Swift for on-device mobile, Python/TypeScript for cloud/Lambda.
  • Classifier performance (intent routing accuracy) directly impacts user experience; test with representative queries before production.
  • Context window and cost scale with conversation length; plan for session truncation or archival in long-running chatbots.
  • Tool definitions and MCP server availability differ by runtime; validate tool coverage for your use case before migration.
  • Grounded agent pattern requires two LLM calls per turn (gatherer + presenter); cost and latency implications are non-trivial.

When to avoid it — and what to weigh

  • Standardized single-agent chatbot — If you need only one LLM endpoint with no routing logic, the framework's multi-agent overhead adds unnecessary complexity.
  • Real-time low-latency requirements (<100ms) — Classifier routing and inter-agent context synchronization add latency; not suitable for sub-100ms SLA systems.
  • Closed, proprietary LLM integrations only — Framework focuses on open APIs and Bedrock; if locked to a single vendor's closed SDK, integration effort increases.
  • Requirement for HIPAA/FedRAMP-certified agents — No explicit certification data provided; requires security review before health/federal use.

License & commercial use

Apache License 2.0 (Apache-2.0). Permissive open-source license permitting commercial use, modification, and distribution under Apache 2.0 terms.

Apache 2.0 is a permissive OSI license that explicitly allows commercial use, derivative works, and closed-source applications provided the license text and notice are retained. No royalties or restrictions on profit-bearing products. Recommended for commercial buyers to review the LICENSE file and ensure internal compliance teams audit dependencies.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

SignalAssessment
MaintenanceActive
DocumentationStrong
License clarityClear
Deployment complexityLow
DEV.co fitStrong
Assessment confidenceHigh
Security considerations

Framework itself does not claim specific security certifications. LLM responses are not inherently safe; GroundedAgent reduces hallucination but does not guarantee correctness. Credentials (API keys, AWS IAM) must be managed by the deployment environment, not the framework. Tool invocation introduces remote execution risk—validate MCP servers and HTTP endpoints. On-device Swift runtime avoids cloud transmission but does not prevent local data exfiltration. Recommend security review before handling PII, health data, or financial information.

Alternatives to consider

LangGraph (LangChain)

State-machine-based agent orchestration with strong Python ecosystem; better for complex DAGs, worse for simple classifier-driven routing.

Crew AI

Role-based multi-agent framework with built-in hierarchies; more opinionated, less flexible for custom classifier logic.

AWS Lambda with raw boto3 + step functions

Lower abstraction, more control; requires manual orchestration and context management; no multi-language support.

Software development agency

Build on agent-squad with DEV.co software developers

Explore Agent Squad's documentation, try a quick-start example, and assess fit for your multi-agent use case. Contact us to discuss cloud deployment or mobile integrations.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

Related on DEV.co

Explore the category and the services that help you build with it.

agent-squad FAQ

Does Agent Squad run on-device without cloud APIs?
The Swift runtime can use local models (via OpenAI-compatible APIs on-device), but requires an LLM. Python/TypeScript typically call cloud LLM APIs. Grounded agent and MCP tools can run locally in Swift; Python/TypeScript tools depend on integration.
How does the classifier choose which agent to use?
By default, an LLMClassifier analyzes agent descriptions and conversation history to pick the best match. Custom classifiers (rule-based, ML) can be plugged in. Classifier performance directly affects user experience.
What is the cost overhead of the GroundedAgent pattern?
Two LLM calls per turn (gatherer analyzes tools, presenter writes response). Cost is roughly 2× a single-agent approach. Saves on hallucination errors and refetches, which can offset in high-accuracy domains.
Is Agent Squad suitable for healthcare or financial applications?
Not without thorough security and compliance review. No HIPAA/FedRAMP certification claimed. LLM hallucination, tool errors, and data residency must be validated. GroundedAgent reduces hallucination but does not guarantee correctness.

Work with a software development agency

Adopting agent-squad is usually one piece of a larger software development effort. As a software development agency, DEV.co provides software development services and web development expertise — pairing senior software developers and web developers with your team to design, build, and operate ai frameworks software in production.

Ready to orchestrate intelligent agent workflows?

Explore Agent Squad's documentation, try a quick-start example, and assess fit for your multi-agent use case. Contact us to discuss cloud deployment or mobile integrations.