The agent landscape, read backwards from the work.

Every map of this market starts with vendors. Three hundred-plus companies across eighteen categories, each homepage promising to run your business. This one starts from the other end, with the task a person actually has, and asks which layer of the stack is even involved. Below: 140 tools, twelve layers, and an honest account of where our own product sits.

140
tools indexed here, across twelve layers
307
agent startups tracked across 18 categories
VC Maps, 2026
40%
of agentic AI projects forecast cancelled by end of 2027
Gartner
95%
of generative AI pilots showing no measurable P&L return
MIT Project NANDA

A supply problem the market keeps treating as a demand problem.

Capability is not the bottleneck any more. Naming the work is.

Everything, for everyone

Read ten homepages in this directory and you will find the same sentence. Agents that run your business. The category is genuinely differentiated on the inside and almost identical on the outside.

The buyers are not founders

A founder reads a pricing page and infers a workflow. A bookkeeper, a paralegal or a store manager cannot. Most people have never seen their own job written down as a list of tasks, let alone scored.

The failures are scoping failures

Gartner expects four in ten agentic projects to be cancelled by 2027. MIT found almost no measurable return across generative pilots. Very little of that is the model being incapable. It is nobody naming the job first.

The stack, ordered by distance from the human.

Most market maps group by company type. This one groups by how far a layer sits from the person with the task, because that is what decides who has to understand it.

01Generalist agentsOne chat box, any task. What a normal person means when they say "an AI that does things". Closest to the human, and the only layer most people will ever see.
02Vertical agentsOne job, done end to end, sold as an outcome rather than a seat. Support resolution, legal review, coding, outbound. The layer with the clearest revenue.
03Enterprise platformsAgents shipped inside the system of record you already bought. Distribution beats capability here, which is why the incumbents are doing well.
04BuildersAssemble an agent without writing code. The most crowded layer, the one every founder thinks of first, and the one where integration count is the actual product.
05FrameworksCode-first libraries for people who want the loop in their own repo. Free, open, and where most of the design patterns are set.
06Tools and authHow an agent touches other software, and how it borrows a specific user’s permissions to do it safely. Boring, invisible, and the hardest thing on this page to build.
07Web, data and docsFetching, crawling, extracting and structuring the outside world so an agent can read it. Includes the new category of documentation written for machines.
08Memory and knowledgeWhat the agent retains between runs, and how it retrieves what the company knows. Vector stores, graphs and the search layer over internal systems.
09Runtime and sandboxesSomewhere safe for generated code and browser sessions to execute. Rented by the second.
10Eval and observabilityWhether the thing worked, why it failed, and what it cost. The layer enterprises discover exists on the day their pilot stalls.
11Models and routingThe raw capability, plus the gateways that pick between them. Furthest from the human and the only layer where progress is genuinely exponential.
12Discovery layerWhere buyers go to find any of the above. Currently indexed by vendor, alphabetically. This is the layer we are competing in.

The directory.

Entry price as published by the vendor, checked July 2026. Blank means the vendor does not publish one. This market reprices constantly, so treat every number as a starting point rather than a quote.

140 of 140

Generalist agents

Browses, runs code and completes multi-step tasks inside the chat everyone already has.
Everyone
From $20/mo
Long-context assistant with computer use, plus Claude Code for engineering work.
Everyone
From $20/mo
Google’s assistant, with Project Mariner for browser control and deep Workspace reach.
Everyone
From $20/mo
Autonomous agent that plans and executes long tasks in its own virtual machine.
Prosumers
From $20/mo
Super-agent bundling research, calls, slides and video generation on a credit meter.
Prosumers
From $24.99/mo
Answer engine with agentic research and shopping actions.
Everyone
From $20/mo
European assistant with free Gmail and Calendar hooks.
Everyone
Free tier

Vertical agents

Customer-facing support agent charged per resolution, not per seat.
Enterprise support
Outcome-based
Enterprise support agent built to close tickets end to end without escalation.
Enterprise support
Custom
Fin
Support agent across voice, chat and email. Publishes a 76% average resolution rate.
Support teams
Per resolution
Legal agent for contract review, case research, diligence and drafting.
Law firms
Custom
Personal injury case preparation and demand package generation.
Injury firms
Custom
Clinical-safety-focused healthcare agents for patient-facing calls.
Providers
Custom
Clay
Go-to-market data enrichment and research agents feeding outbound.
Sales teams
From $149/mo
11x
Digital SDR that sources, researches and sequences outbound autonomously.
Sales teams
Custom
Outbound sales agent sold explicitly as a headcount replacement.
Sales teams
Custom
Ten-plus marketing agents running SEO, Reddit, X, LinkedIn and outreach continuously.
Solo founders
From $99/mo
Autonomous software engineer taking tickets through the full development cycle.
Eng teams
From $20/mo
AI-native IDE. The default agentic coding surface for most professional teams.
Developers
From $20/mo
Terminal-native coding agent with subagents and multi-agent coordination.
Developers
From $20/mo
Agent mode that iterates on issues inside the repository you already use.
Developers
From $10/mo
Agentic IDE built around multi-file reasoning.
Developers
From $15/mo
Prompt to deployed application. Reported $75M ARR.
Non-engineers
From $25/mo
v0
Vercel’s generative interface and app builder.
Product teams
From $20/mo
Builds and hosts full-stack applications from a description.
Non-engineers
From $25/mo
In-browser full-stack app generation.
Non-engineers
From $20/mo
Open-source software engineering agent. Claims 87% of bug tickets closed same day.
Eng teams
Open source
Design and ship chat and voice agents for customer-facing channels.
CX teams
From $60/mo

Enterprise platforms

Agents inside the CRM. Reported past $540M ARR across 18,500 customers.
Salesforce shops
Custom
Build agents that live in Teams and Microsoft 365.
M365 shops
From $200/mo
Agents across IT, HR and customer workflows on the existing platform.
Enterprise IT
Custom
HR and finance agents embedded in the system of record.
HR and finance
Custom
Domain agents and several hundred prebuilt tools with enterprise governance.
Enterprise IT
Custom
Collaborative agents plus Joule Studio for building them.
SAP shops
Custom
Ontology-grounded autonomous operations. The most opinionated take on agent grounding.
Large enterprise
Custom
Work search and assistants over everything the company knows. Reported $300M ARR.
Large enterprise
Custom
Employee support agents for IT and HR service desks.
Enterprise IT
Custom
Agentic service desk automation across IT, HR and customer service.
Enterprise IT
Custom
Enterprise agent platform spanning contact centre and internal workflows.
Enterprise
Custom
Conversational and voice agents for large contact centres.
Contact centres
Custom
Resolution agents inside the helpdesk already in place.
Support teams
Custom
Enterprise automation with 1,200-plus connectors and an AI recipe builder.
Enterprise ops
Custom
Universal automation cloud with agent capabilities layered on.
Enterprise ops
Custom

Builders

Agent teams where each member runs on a different model, with a visible routing receipt. Ours.
Individuals
Free tier
n8n
Visual workflow automation with roughly 70 AI nodes. Self-hostable and free that way.
Technical ops
Free self-host
The default automation layer, now with Agents, Tables and Canvas bolted on.
Everyone
From $29.99/mo
Make
Visual scenario builder over 3,000-plus apps, materially cheaper than Zapier at volume.
Ops teams
From $9/mo
Build "AI employees" in plain English. Claims 5,000-plus integrations.
Non-technical
From $19.99/mo
Teams of specialised agents sold as an AI workforce, strongest in sales.
GTM teams
From $29/mo
Visual canvas for multi-agent orchestration, aimed at technical builders.
Technical ops
From $37/mo
No-code agent platform with straightforward deployment.
Non-technical
From $20/mo
Agents that drive real browsers and desktop apps, and build their own interfaces.
Lean teams
$200/user/mo
Describe a task, get a deterministic Agent App that runs on a schedule without a model in the loop.
Solo builders
Free tier
Dify
Open-source LLMOps with a visual workflow builder. Very large community.
Developers
Open source
Drag-and-drop agent construction on top of LangChain.
Developers
Open source
Low-code builder for agentic and retrieval applications.
Developers
Open source
Coze
ByteDance’s agent builder with a large plugin ecosystem and a generous free tier.
Everyone
Free tier
MIT-licensed open alternative to Zapier and Make.
Technical ops
Open source
Natural language treated as the programming language for agents.
Prosumers
Free tier
Visual backend builder for AI workflows and agent endpoints.
Developers
From $25/mo
Enterprise-oriented no-code builder with governance and on-prem options.
Enterprise
From $199/mo
Enterprise AI orchestration with a no-code agent layer.
Enterprise
Custom
Visual programming environment for agent graphs, from Ironclad.
Developers
Open source

Frameworks

The foundational Python and JS library most agent code still touches somewhere.
Developers
Open source
Graph-based orchestration for stateful, multi-actor agent workflows.
Developers
Open source
Role-based multi-agent crews. Claims adoption across most of the Fortune 500.
Developers
Open source
Microsoft’s conversation-driven multi-agent framework.
Developers
Open source
Data framework with 160-plus connectors for retrieval and agent workflows.
Developers
Open source
Lightweight production agent runtime, the successor to Swarm.
Developers
Open source
Build agents on Claude with custom tools, hooks and subagents.
Developers
Open source
Code-first agent toolkit, tuned for Gemini but model-agnostic.
Developers
Open source
TypeScript-first agent framework from the Gatsby team.
JS developers
Open source
Type-safe agents with a FastAPI-like developer experience.
Python devs
Open source
Agno
High-performance multi-modal agent runtime.
Developers
Open source
HuggingFace library where agents write Python instead of emitting JSON.
Developers
Open source
Formerly MemGPT. Stateful agents with first-class long-term memory.
Developers
Open source
DSPy
Stanford framework for programming and optimising language model pipelines.
Researchers
Open source
Microsoft’s enterprise agent SDK, strongest for .NET teams.
Enterprise devs
Open source
Production-oriented orchestration framework from deepset.
Developers
Open source

Tools and auth

Tool layer giving agents over 1,000 toolkits with managed connected accounts.
Developers
Free tier
MCP runtime built around per-user OAuth delegation and audited authorisation.
Developers
Free tier
10,000-plus prebuilt actions across 200-plus connectors with managed auth.
Product teams
Custom
Developer infrastructure for custom API integrations, around 900 of them, self-hostable.
Developers
Free tier
Embedded integrations with ActionKit exposing 1,000-plus tools in one call.
SaaS vendors
Custom
Roughly 2,700 APIs behind a low-code workflow layer. Acquired by Workday.
Developers
Free tier
Unified API across HRIS, ATS, CRM and accounting categories.
SaaS vendors
Custom
MCP
The tool-connection standard, created by Anthropic and donated to the Linux Foundation.
Everyone
Open standard
A2A
Google-originated agent-to-agent protocol with broad vendor backing.
Platform teams
Open standard

Web, data and docs

Turns any site into clean model-ready markdown, with crawling and extraction.
Developers
From $16/mo
Exa
Neural search API built for retrieval by machines rather than people.
Developers
Usage-based
Search and extraction API designed specifically for agent grounding.
Developers
From $30/mo
One API for scraping, crawling, parsing, schema extraction and brand intelligence.
Developers
From $25/mo
Managed cloud browsers for agents, with Stagehand for self-healing actions.
Developers
Usage-based
Open-source browser agent framework. Reports 89% on the WebVoyager benchmark.
Developers
Open source
Open-source headless browser API, self-hostable.
Developers
Free tier
Vision-model browser automation that reads pages from screenshots.
Developers
Free tier
Open-source crawler producing clean markdown for retrieval pipelines.
Developers
Open source
Marketplace of prebuilt scrapers and actors, callable as agent tools.
Ops and devs
From $39/mo
Repositioned as a context layer, moving source systems into a queryable store agents can reason over.
Data teams
From $10/mo
Documentation written to be read by agents, with an agent-readability score and MCP server.
Product teams
Free tier
Turns text corpora into knowledge graphs and surfaces the structural gaps in them.
Researchers
From €12/mo

Memory and knowledge

Mem0
Memory layer for agents. Named as the memory provider for the AWS Agent SDK.
Developers
Free tier
Zep
Temporal knowledge graph memory, built on the open-source Graphiti library.
Developers
Free tier
Fully managed vector database. The enterprise default at scale.
Developers
Free tier
Open-source vector database with native multi-tenancy and hybrid search.
Developers
Open source
Rust-based high-performance vector search, self-hostable.
Developers
Open source
Lightweight embedded vector store, the usual choice for prototypes.
Developers
Open source
Vector search inside Postgres, with no additional infrastructure at all.
Developers
Open source
Embedded vector database on a columnar format, good for local and edge use.
Developers
Open source

Runtime and sandboxes

E2B
Secure cloud sandboxes for running agent-generated code.
Developers
Free tier
Fast-provisioning sandboxes aimed at agent workloads.
Developers
Free tier
Serverless compute billed by the second, widely used for agent tool execution.
Developers
Usage-based
Model inference infrastructure with dedicated GPU capacity.
ML teams
Usage-based

Eval and observability

Tracing, evaluation and monitoring across the agent lifecycle. Framework-agnostic.
Developers
Free tier
Open-source LLM observability, acquired by ClickHouse. Large paying base.
Developers
Open source
Evaluation and monitoring used by Notion, Stripe and Vercel.
Product teams
Free tier
Open-source tracing on OpenTelemetry, vendor-neutral by design.
ML teams
Open source
Open-source observability and gateway, Rust-based for throughput.
Developers
Free tier
AI gateway with routing, caching and dozens of prebuilt guardrails.
Platform teams
Free tier
Session replay and failure detection for agent runs, two lines to install.
Developers
Free tier
Evaluation and real-time hallucination detection for production systems.
Enterprise
Custom
Automated testing and failure analysis for enterprise deployments.
Enterprise
Custom
Red teaming and vulnerability scanning for prompts and agents.
Security teams
Open source

Models and routing

One API across 200-plus models, with routing and fallback. What we build on.
Developers
Usage-based
Open-source provider abstraction over 100-plus models with load balancing.
Developers
Open source
Claude models with long context, computer use and strong tool calling.
Developers
Usage-based
GPT and Codex families plus the hosted agent and deep research surfaces.
Developers
Usage-based
Gemini models with browser control and very large context windows.
Developers
Usage-based
Groq
Ultra-low-latency inference, useful when an agent has to answer in real time.
Developers
Usage-based
Open-model inference and fine-tuning at scale.
Developers
Usage-based
Fast hosted inference for open and custom models.
Developers
Usage-based
MIT-licensed frontier-adjacent open models, trained remarkably cheaply.
Developers
Open weights
Qwen
Apache-licensed Alibaba models, the most-derived open family on HuggingFace.
Developers
Open weights
Open-weight family licensed for commercial use.
Developers
Open weights

Discovery layer

Vendor-indexed marketplace across 70-plus categories. Sells six ad slots and partner spots.
Buyers
Ad-funded
Directory of 1,300-plus agents with a 30% affiliate programme on paid placements.
Buyers
Ad-funded
The largest general AI tool index. Broad, shallow, monetised by placement.
Buyers
Ad-funded
G2
Review-led software discovery. Owns the category-level buying keyword.
Buyers
Ad-funded
Investor-facing market map tracking 307 agent startups across 18 categories.
Investors
Free
HATL
Indexed by occupation and task rather than by vendor. This page. Ours.
Workers
Free

Method. Figures are vendor-published or press-reported and have not been independently verified. Funding, revenue and star counts move weekly. Nothing here is a paid placement. Where a tool spans layers it is filed under the one its own homepage leads with.

Where our own product stands, stated plainly.

A directory that ranks tools and quietly owns one is worth nothing. So here is the position, including the part that does not flatter us.

Where it cannot win

Integration breadth.
  • Composio lists over 1,000 toolkits. Nango is around 900. Pipedream connects roughly 2,700 APIs. Make lists 3,000-plus apps. Lindy claims 5,000-plus integrations.
  • Keimodel ships one read connector, for Shopify, plus three delivery channels.
  • That gap is three orders of magnitude and it is bought with capital and years, not cleverness.
  • Treating it as a roadmap item is how a small team spends two years losing a race it already lost.

Where it can win

The model layer, and the first thirty seconds.
  • Routing across 200-plus models with a visible receipt for why each was picked. Almost every builder in this directory hides the model choice or hard-codes it.
  • Teams where each member runs on a different model and hands off. Sold elsewhere as an enterprise feature.
  • Most useful work needs no integration at all. Research, drafting, analysis, planning and reporting all produce a deliverable from a prompt and a web search.
  • It is the only tool on this page attached to a demand engine that knows what the visitor’s job is before they arrive.

What it therefore is

The bench, not a contender.
  • Keimodel runs a real sample of your task in the page, free, before you have signed up for anything.
  • Anything needing a live system of record gets routed out to the vendors above, and that routing is the business.
  • Keimodel is never ranked against them here. It is the try-it surface, labelled as ours.
  • Which tasks people try, and which routed clicks convert, is the signal for the only connectors worth ever building.

What we are building toward.

The short version: everyone here sells a hammer to people who cannot name their nail. The index of nails is the asset.

The unit is the task, not the tool.

Every directory in this market has the vendor as its atom, which is why they all converge on the same seventy alphabetised categories and the same six advertising slots. They start from supply because supply is what emails them.

Our atom is the task. 1,016 occupations and roughly 19,000 task statements from the public-domain O*NET database, each scored for how much of its output an agent can actually produce, each mapped to the tools that could produce it. A vendor list is copyable in a weekend. A calibrated map of work to capability is not.

Directories index who is selling. We index what needs doing.

We own the entry point nobody wants.

Risk quizzes tell you your job is 43% exposed and stop there. Agent builders open a blank canvas and assume you already know what to build. The distance between those two screens is where every abandoned pilot lives, and it is unclaimed because it is unglamorous and needs a labour dataset rather than a model.

Three page types scale programmatically off data we already hold: the occupation, the individual task, and the tool. The fourth type, task crossed with tool, is where the commercial intent is.

The demo is the moat.

No comparison site can execute. We can. A visitor sees their job broken into tasks, then watches one of those tasks actually run, on a real model, in the page, before signing up for anything. That is not a feature a directory can add later. It requires owning a runtime, which is what Keimodel is for.

This also resolves the conflict of interest cleanly. Keimodel does not compete in the rankings. It is the bench where you try the work, and the ranked vendors are where you go to do it at scale.

Three revenue layers, in the order they become possible.

First, referral on routed clicks. It needs no business development and costs nothing to turn on. Second, placement against specific task clusters, which is worth far more than a banner because the intent is legible: this visitor is a bookkeeper who just learned that invoice reconciliation is 88% automatable. Third, and the real one, the demand data.

Which tasks, in which occupations, people are trying to automate this month is a dataset nobody else can assemble. Every vendor sees only its own funnel. We see the question before the tool is chosen. That is worth more to Sierra or Lindy than a logo slot, and it compounds.

The risk, named before someone else names it.

This is a media and marketplace hybrid, and both halves are exposed. The tool layer will consolidate, and AI answer engines are already absorbing the traffic that classic directories lived on.

The hedge is structural rather than hopeful. Task-level structured data is the format answer engines cite rather than replace, and the task graph survives consolidation intact. If the vendor list collapses from 300 companies to 12, the question of what a paralegal’s day is made of does not change at all. Only the right-hand column does.