FastAgency

Ship AG2 multi-agent workflows from prototype to console, web app, or distributed service in minimal code.

Source repo
ag2ai/fastagency
Stars
★ 546
Last updated
11mo ago
License
Apache-2.0
Primary language
Python

At a glance

Works with
Portable with changesOpenAI API
You'll need
Python 3.10–3.12pipCookiecutterDocker (optional, for image builds)Fly.io account and FLY_API_TOKEN (optional, for deployment)OPENAI_API_KEYShell / CLINetwork accessLocal filesystem
Typical use
Teams that already validated an AG2/AutoGen multi-agent flow in a notebook and want a presentable web chat app quickly.
Main limitation
Runtime lock-in: AG2 (formerly AutoGen) is the only supported runtime; others are future plans.
Website
fastagency.ai/latest ↗ set on GitHub
Maintenance
No commits for 6+ monthslast commit 2025-12-09latest release v0.10.2 (2025-12-09)

What does this agent do, and when should you use it?

FastAgency is a deployment framework for multi-agent workflows written in AG2 (formerly AutoGen), built to move notebook prototypes into production. A single workflow definition runs unchanged behind ConsoleUI for terminal interaction or MesopUI for a web chat interface. Networking is handled by chainable adapters: FastAPIAdapter serves workflows on an ASGI server with multiple workers, while NatsAdapter uses the NATS.io message broker via FastStream for distributed setups spanning machines and datacenters. The project also ships a Tester class for CI regression testing of agent behavior and a CLI for orchestrating workflows from the terminal. External services can be wired in by importing an OpenAPI specification. AG2 remains the only supported runtime today.

A developer registers a function as a named workflow inside a Workflow object with the @wf.register decorator; the function receives a UI object and a params dict, calls ui.text_input to collect user input, defines agents with AG2's ConversableAgent and LLMConfig, runs a conversation, and returns the result through ui.process. The same workflow can be bound to ConsoleUI for terminal use or MesopUI for a Mesop-powered web app. For deployment, FastAPIAdapter exposes the workflow through FastAPI/ASGI with multiple workers, and NatsAdapter can be layered on to scale over NATS.io message brokers. OpenAPI specs can be imported to give agents access to external APIs, the Tester class generates and runs interaction tests, and the CLI launches, parameterizes, and monitors workflows.

  1. Teams that already validated an AG2/AutoGen multi-agent flow in a notebook and want a presentable web chat app quickly.
  2. Developers who need the same workflow available both as a terminal debugging tool and as a web-facing product.
  3. Engineering teams deploying agent services as multi-worker FastAPI/ASGI processes inside existing operations tooling.
  4. Organizations scaling agent workloads across machines or datacenters and willing to run a NATS.io broker.
  5. Teams wanting automated regression tests of multi-agent conversations in CI after prompt or code changes.
  6. Developers adding third-party REST APIs such as weather or payment services to an agent workflow.

How do you install or deploy this agent?

Install Cookiecutter, then generate a project from the official template, which creates the folder layout, a default workflow, dependencies, and a devcontainer.

How do you use this agent?

After generating the project, export your OpenAI API key, reopen the folder in the VS Code devcontainer, confirm the setup with pytest, then start the Mesop UI with gunicorn on Linux/macOS or waitress on Windows. The workflow itself lives in workflow.py and is registered with @wf.register.

What are this agent's strengths and limitations?

Pros
  • One programming interface: the same workflow runs behind ConsoleUI and MesopUI without rewriting code.
  • Clear deployment tiers: FastAPIAdapter for multi-worker ASGI serving, NatsAdapter on NATS.io for cross-machine scaling.
  • A built-in Tester class lets you put multi-agent interaction tests into CI.
  • Low-friction external API integration: import an OpenAPI spec and connect it to agents.
  • Cookiecutter template plus devcontainer, with scripts for Docker builds, Fly.io deploys, and GitHub Actions.
Limitations
  • Runtime lock-in: AG2 (formerly AutoGen) is the only supported runtime; others are future plans.
  • You must supply LLM credentials such as OPENAI_API_KEY and pay the model provider per usage.
  • Production deployment pulls in FastAPI, NATS.io, Docker, and Fly.io, a heavier lift than using a plain library.
  • UI choices are limited to ConsoleUI and MesopUI, with the web path tied to the Mesop framework.
  • Distributed mode requires operating your own NATS.io message broker.

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
FastAgency This agent 47 · Major gaps — ★ 546 11mo ago Python OpenAI API
DeterminFlow 73 · Some gaps Desktop appFree + model costs ★ 979 2mo ago Python —
DevAll 54 · Major gaps CLIFree + model costs ★ 35k 15d ago Python —
Open Swarm Insufficient evidence — ★ 826 2d ago JavaScript Claude API

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
47/ 100 5-point scale 2.4 / 5
Trust 10/29
Reliability 8/14
Adaptability 10/18
Convention 9/18
Effectiveness 7/13
Verifiability 3/8
Why each dimension lost points
Trust10 / 29 · 1.7/5

The trust dimension is weak overall. Least privilege: the default Mesop security policy allows script sources from cdn.jsdelivr.net and iframe parents from fastagency.ai, a fairly broad default allowance with no documented tightening on demand; points deducted. User confirmation: workflows interact with users via ui.text_input/multiple_choice, but there is no explicit confirmation or approval mechanism for external API calls or function execution (suggested_function_call/function_call_execution); 1 point only. Data-flow transparency: the README mentions OpenAPI integration and external APIs, but does not disclose data flows to third-party LLMs/APIs; points deducted. Sensitive data handling: examples pass OPENAI_API_KEY via environment variables and GitHub Secrets, which is reasonable, but the repository shows no key redaction, log redaction or PII handling notes; 1 point only. Dependency security: pyproject constrains versions for most dependencies and there are CI configurations such as CodeQL, dependency-review, bandit, semgrep and detect-secrets, giving fairly sufficient evidence for 2 points; no known-vulnerability scan results are provided, so no full marks. External effects: the framework calls external LLMs and APIs, but no side-effect boundaries or sandboxing are documented; points deducted. Rollback: no rollback, undo or transactional recovery mechanism is evident; 0 points. Source attribution: LICENSE is Apache-2.0 and pyproject lists author ag2ai, but publisher identity is unverified, and package.json declares license ISC, inconsistent with Apache-2.0; points deducted.

Reliability8 / 14 · 2.9/5

The README's feature description (ConsoleUI/MesopUI, FastAPI/NATS adapters, Tester class, CLI) is broadly consistent with pyproject optional dependencies and source structure, and example code corroborates the e2e files; 2 points. Dependency availability: dependencies are explicitly listed in pyproject with version constraints and optional dependency groups are clear; 2 points; however, some dependencies (e.g. fastapi-code-generator==0.5.2, agentwire==0.1.4) are older or niche packages with availability risk, so no full marks. Failure messages: the Mesop home page only logs exceptions and shows generic error text, and in the retry logic the except branch sets available_workflows_exception to False, a suspected logic defect; error messages offer limited help to users; points deducted.

Adaptability10 / 18 · 2.8/5

The README clearly targets developers using AG2 and covers console, Web, REST and NATS distributed scenarios; audience and scenarios are well described; 2 points. Capability boundaries: it explicitly states that only the AG2 runtime is currently supported and lists supported UI and network adapters; boundaries are clear; 2 points. Trigger precision: workflows are registered via @wf.register with name/description, but there is no documentation of trigger conditions, ambiguity resolution or selection logic; points deducted. Environment fit: supports Python 3.10-3.12, distinguishes Windows/Linux start commands (gunicorn/waitress), and provides devcontainer and Docker scripts; environment fit is good; 2 points.

Convention9 / 18 · 2.5/5

The README is clearly structured with sections on features, quick start, deployment and future plans, and links to the documentation site; 2 points. Install notes: provides cookiecutter scaffolding, pip install, environment variable setup and pytest verification steps, fairly complete; 2 points. Naming stability: in the examples multiple_choice_single_workflow is defined repeatedly (overwriting the previous function), and registered names differ from function names, indicating naming confusion; points deducted. Examples and FAQ: the README and e2e directory provide multiple runnable examples, but there is no standalone FAQ; 2 points. Known limitations: only the current AG2-runtime-only support is mentioned; limitations and known issues are not systematically listed; points deducted. License: LICENSE is the full Apache-2.0 text and pyproject classifiers are consistent; 2 points; however, package.json declares license ISC, an inconsistency already penalized under source attribution. Versioning and changelog: pyproject uses a dynamic version (__about__.py) and no CHANGELOG file is evident; points deducted. Maintenance responsibility: author email, Discord and Issue links exist, but publisher identity is unverified and no explicit maintenance commitment or SLA is evident; points deducted.

Effectiveness7 / 13 · 2.7/5

The framework provides a unified UI interface, CLI, Tester class and deployment scripts, and outputs can be used directly to build applications; 2 points. Marginal value: it provides UI and deployment abstractions on top of AG2, reducing boilerplate, with clear value; 2 points. Cost benefit: it introduces many optional dependencies (mesop, fastapi, nats, firebase, etc.) and some dependency versions are old, so deployment and maintenance costs are not low; points deducted.

Verifiability3 / 8 · 1.9/5

Feature claims in the README can be mapped to source files (e.g. MesopUI, Workflow registration), but marketing-style statements about performance and production readiness lack verifiable evidence; points deducted. Cross-source corroboration: README, pyproject, e2e examples and test files corroborate some features, but test samples are limited; 1 point only. Fact-inference separation: the documentation does not clearly distinguish verified facts from plans (Future Plans are labeled), but some capability descriptions are vague; points deducted.

Risks and how to mitigate them
  • Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
  • The default Mesop security policy allows scripts from cdn.jsdelivr.net and iframes from fastagency.ai; review and tighten before deployment.
  • The framework calls external LLMs and third-party APIs, but data-flow disclosure and user confirmation are not documented; add controls when handling sensitive data.
  • package.json declares license ISC, inconsistent with Apache-2.0; verify.
  • Publisher identity is unverified; do not infer reliability from the brand.
  • Example code contains duplicate function definitions and naming inconsistencies; review before reuse.
  • No rollback or transactional recovery mechanism is evident; design your own for production.
Evidence confidence: Low Reviewed Nov 16, 2026 Reviewed revision f353682dffc6
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/fastagency.svg)](https://followagents.com/en/agents/fastagency)
<a href="https://followagents.com/en/agents/fastagency"><img src="https://followagents.com/badge/fastagency.svg" alt="FARS review"></a>

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

  1. v0.10.2
    Release notes excerpt
    What's Changed * Bump version to 0.9.9 (incl. remove Python 3.9 support) by @marklysze in https://github.com/ag2ai/fastagency/pull/736 * FastAgency + AG2 version bump 0.10.2 by @marklysze in https://github.com/ag2ai/fastagency/pull/750 Full Changelog: https://github.com/ag2ai/fastagency/compare/v0.9.7...v0.10.2
    View on GitHub ↗
  2. v0.9.7
    Release notes excerpt
    What's Changed * Bump version to 0.9.7 by @marklysze in https://github.com/ag2ai/fastagency/pull/732 Full Changelog: https://github.com/ag2ai/fastagency/compare/v0.9.6...v0.9.7
    View on GitHub ↗
  3. v0.9.6
    Release notes excerpt
    What's Changed * Bump version to 0.9.5 by @marklysze in https://github.com/ag2ai/fastagency/pull/729 * Bump version to 0.9.6 by @marklysze in https://github.com/ag2ai/fastagency/pull/730 Full Changelog: https://github.com/ag2ai/fastagency/compare/v0.9.4...v0.9.6
    View on GitHub ↗

FAQ

Is FastAgency open source?
Yes. The source code is on GitHub (ag2ai/fastagency) under the Apache-2.0 licence.
Is FastAgency safe to use?
FollowAgents' FARS review (Nov 16, 2026) scored FastAgency 10/29 on trust and safety. The review did not find these safety controls in the source: rollback or recovery path. Before giving it real accounts or data: Back up first, or work on a git branch or snapshot, so its changes can be undone.
Is FastAgency still maintained?
The GitHub repository has had no commits since 2025-12-09, more than six months ago. That can mean the project is finished or abandoned; check its issues before relying on it. The latest release is v0.10.2 (2025-12-09).
Who makes FastAgency?
FastAgency is published on GitHub by the account ag2ai. FollowAgents has not verified which person or company is behind that account.
What is FastAgency's official website?
The project's GitHub repository lists https://fastagency.ai/latest as its website. That link is set by the repository owner; the source code itself is on GitHub (ag2ai/fastagency).
Was this page helpful?
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts