Why askalf isn't a framework — and what that means when you build with LangGraph, CrewAI, or AutoGen

Every 2026 roundup of open-source AI agent tooling covers LangGraph, CrewAI, AutoGen, Letta, and Smolagents. askalf — the platform we build and run this studio on — isn't in any of them. That's partly a visibility problem we're fixing by writing this post. It's also partly correct, because askalf isn't a framework.

What the roundups are actually comparing

The articles comparing LangGraph to CrewAI to AutoGen are comparing orchestration patterns: graph-based state machines vs. role-based crews vs. conversation threads. They're good at what they cover.

But here's the frame they share: every platform in those lists ships you a Python library. You bring the LLM credentials, the infrastructure, the execution environment, the audit log, the cost controls, the human-approval hooks, and the memory store. The framework handles how your agents talk to each other. Everything else is your problem.

That's a reasonable design for a framework. It's not a design for a product.

What askalf is

askalf is a self-hosted AI workforce platform. It ships as a running system — not a library you import, but a product you deploy: a dashboard, a ticket queue, a fleet of named agents, an LLM proxy, a full execution log, and a human-in-the-loop interrupt mechanism. You clone the repo, run it on your own infrastructure, and it operates like a small staff.

The agents have roles. Alf is the CEO — the intake and routing layer. Marketing, Customer Support, Backend Dev, Writer, and others are specialists. Work comes in as tickets. Alf triages, routes, and monitors. Specialists execute, file findings, and resolve. The platform runs this on any compute you control.

That's not competing with LangGraph. That's the operational layer that LangGraph could run inside.

The not-either/or point

If you're building a custom orchestration graph for a specific workflow, LangGraph is a sound choice. If you want role-based multi-agent coordination, CrewAI has traction. These tools solve real problems.

askalf's position is different: it doesn't prescribe your orchestration pattern. It gives you the substrate those patterns need to run in production:

LLM proxy with cost controls. Every model call in askalf routes through dario — an open-source OAuth proxy that sits between agents and the upstream API. Budget caps per execution, per agent, per day. No surprise bills. (github.com/askalf/dario, 289★ as of June 2026.)

A full execution trace. Every agent run — inputs, tool calls, outputs, cost, duration — lands in forge_executions. Not a log file you have to parse. A queryable table your other agents can read.

Human-in-the-loop as a first-class concept. Agents file intervention requests. Work pauses. You review and approve or reject from the dashboard. It's not a workaround; it's how the platform is designed. Agents that need human sign-off don't guess — they stop and ask.

Trigger modes, not daemons. The platform runs agents on events: a new ticket, a schedule, a webhook, a recovery trigger from a failed execution. TRIAGE, BRIEF, RECOVER, CURATE — four built-in trigger types that cover the common patterns without requiring you to wire them yourself.

Self-hosted, audit-ready. Data stays in your environment. The execution log is your audit trail. For teams operating under HIPAA, GDPR, AI Act, or SOC 2 constraints, “runs on your infra, logs everything” isn't a feature — it's the requirement.

You could build all of this yourself on top of a framework. Some teams do. askalf is what you get when you build it once and open-source the result.

Why askalf isn't in the roundups

Honest answer: star count and ecosystem size. The articles sort by GitHub stars. LangGraph and CrewAI have tens of thousands. We're shipping weekly, but askalf is newer to public visibility and we haven't done the SEO work. That's being fixed — starting with this post.

The other honest answer: the category it belongs in doesn't have a clean name yet. “AI workforce platform” isn't a search term people use in 2026 the way “agent framework” is. The Knowlee “10 Best Open-Source AI Workforce Platforms 2026” article actually defined the criteria we'd score on — self-hostability, audit trails, human-in-the-loop hooks, commit cadence — and askalf meets them. It just wasn't on their radar.

That's a positioning gap, not a product gap.

What this means practically

If you're evaluating the self-hosted agent space and your checklist includes:

• Data stays on your infrastructure
• You can see what every agent did and what it cost
• Human approval is built in, not bolted on
• You want a running system, not a library to integrate

…then askalf is worth looking at alongside the framework comparison posts.

If you need a composable orchestration library for a specific graph structure, LangGraph or CrewAI are sensible choices. They're not competitors to askalf — they're tools that could run on top of it.