Research reportai-agents
The State of AI Agents 2026 — A Data Report on 51 Researched Tools
An analysis of 51 researched AI agents across 7 categories: open-source rates, API availability, deployment models and pricing, from our own listings.
A sourced, data-driven landscape report — built from The Agents Index’s quality-gated listings corpus and designed to be cited by developers, technical buyers, and AI systems evaluating which agent to build on. Every figure below is computed directly from the structured listings; the generator that produces this page is checked into the repository at scripts/generate-agents-report.ts.
Executive Summary
The AI-agent market has splintered into distinct build-vs-buy layers rather than converging on one model. 51 researched agents and agentic tools span 7 categories, from terminal coding agents to no-code sales automation platforms. This report examines the landscape across open-source availability, API access, deployment model, and interface, based on structured listings published by The Agents Index.
Key findings:
- 37% of tools are open source (19/51) — but open-source and self-hostable are not the same claim; see Finding 1.
- 67% expose a documented API (34/51), the baseline for wiring an agent into another system rather than using it standalone.
- Coding agents is the deepest category (16 tools), followed by Agent frameworks (8) and Support agents (6).
- Deployment splits along a Cloud/Self-host/Both line, not a single dominant model — see Finding 3.
- No single pricing model dominates: freemium, usage-based, free, and flat subscription are all in active use across the corpus — see Finding 5.
Landscape Overview: 51 Tools Across 7 Categories
| Category | Group | Tools |
|---|---|---|
| Coding agents | Agents | 16 |
| Agent frameworks | Build | 8 |
| Support agents | Agents | 6 |
| Sales & marketing agents | Agents | 6 |
| Voice agents | Agents | 6 |
| Research & data agents | Agents | 5 |
| Agent platforms | Build | 5 |
Data-Driven Findings
1. Open Source Is Common (37%) But Doesn’t Mean Self-Hostable
Open source: 19/51 tools (37%). Closed source: 32/51 tools (63%).
Within Coding agents specifically — the category our own terminal-agent comparison first flagged this pattern for — 7/16 tools are open source. That earlier piece checked six terminal agents by hand and found license, model portability, and self-hosting were three separate axes that don’t move together; this report confirms the open-source share holds (not just an artifact of which six were picked) across the full 16-tool category.
Interpretation: “Open source” answers a licensing question, not a deployment one — cross-check the Deployment column in the full table below before assuming an open-source agent runs anywhere but the vendor’s own cloud.
2. API Access Is the Baseline, Not the Differentiator (67%, 34/51)
Tools with a documented API: 34/51 (67%). Tools without one documented: 17/51 (33%).
Key finding: a majority of the corpus already exposes an API — for a category built to be wired into other systems, that’s expected. The real differentiator for integration decisions is deployment model (Finding 3) and model flexibility (the modelLLM field on each listing), not API presence alone.
3. Deployment: No Single Model Dominates
| Deployment | Tools | Share |
|---|---|---|
| Cloud | 28 | 55% |
| Both | 10 | 20% |
| Self-host | 9 | 18% |
| Local | 2 | 4% |
| Local + Cloud | 1 | 2% |
| Self-hosted or Cloud | 1 | 2% |
Interpretation: Cloud-only is the largest single bucket but far from universal — a real share of the corpus is self-hostable or offers both, which matters for teams with data-residency or cost-at-scale constraints that a pure-SaaS agent can’t meet.
4. Interface: Where an Agent Lives Varies by Category
| Interface | Tools | Share |
|---|---|---|
| Web | 23 | 45% |
| Library | 9 | 18% |
| Visual | 7 | 14% |
| CLI | 6 | 12% |
| IDE | 4 | 8% |
| API | 2 | 4% |
Interpretation: Coding agents cluster around CLI/IDE interfaces; agent platforms and frameworks lean Web/Visual/Library — a Library-interface tool like LangChain is for developers assembling their own system, a Visual/no-code one is for teams without engineering resources dedicated to the build. An agent’s interface is a strong proxy for who it’s built for.
5. Pricing Models: Freemium Leads, But Usage-Based Is Real
| Pricing Model | Tools | Share |
|---|---|---|
| freemium | 24 | 47% |
| usage-based | 13 | 25% |
| subscription | 7 | 14% |
| free | 7 | 14% |
Interpretation: Unlike a mature SaaS category that converges on flat monthly tiers, agent pricing is still settling — usage-based (per-token, per-minute, per-resolution) pricing is common enough that a flat $/month comparison across tools would misrepresent real cost. This report deliberately does not force tools into price bands for that reason; see Methodology.
6. Capability Tags: The Filter Layer
- autonomous: 32/51 (63%)
- multi-agent: 15/51 (29%)
- self-hosted: 19/51 (37%)
- no-code: 12/51 (24%)
- enterprise: 25/51 (49%)
- browser-use: 4/51 (8%)
- free-tier: 37/51 (73%)
- api: 36/51 (71%)
- open-source: 19/51 (37%)
- claude: 9/51 (18%)
- gpt: 10/51 (20%)
- open-models: 8/51 (16%)
Interpretation: free-tier and autonomous are the two most common capability tags — most agents let you try before paying, and most already operate with some autonomy rather than requiring a human in the loop for every action. Model tags (claude, gpt, open-models) are far from universal — model choice remains a real filter, not a solved commodity.
Full Spec Table (51 Tools)
| Agent | Category | Open Source | API | Deployment | Interface | Entry Price |
|---|---|---|---|---|---|---|
| 11x | Sales & marketing agents | ✗ | ✓ | Cloud | Web | $2,417 |
| Ada | Support agents | ✗ | ✓ | Cloud | Web | Custom |
| Aider | Coding agents | ✓ | ✗ | Self-host | CLI | Free |
| Artisan | Sales & marketing agents | ✗ | ✗ | Cloud | Web | Custom |
| Bland AI | Voice agents | ✗ | ✓ | Both | Visual | $0.11 |
| Bolt.new | Coding agents | ✗ | ✗ | Cloud | Web | $25 |
| Claude Code | Coding agents | ✗ | ✓ | Cloud | CLI | $17 |
| Clay | Sales & marketing agents | ✗ | ✓ | Cloud | Web | $167 |
| Cline | Coding agents | ✓ | ✓ | Self-host | IDE | $9.99 |
| CodeRabbit | Coding agents | ✗ | ✓ | Both | Web | $24 |
| Crescendo | Support agents | ✗ | ✗ | Cloud | Web | From $2.99 |
| Cresta | Support agents | ✗ | ✓ | Cloud | Web | $150,000 |
| CrewAI | Agent frameworks | ✓ | ✓ | Self-host | Library | Free |
| Cursor | Coding agents | ✗ | ✗ | Cloud | IDE | $20 |
| Decagon | Support agents | ✗ | ✗ | Cloud | Web | Custom |
| Devin | Coding agents | ✗ | ✓ | Cloud | Web | $20 |
| Dify | Agent platforms | ✓ | ✓ | Both | Visual | $590 |
| ElevenLabs Conversational AI | Voice agents | ✗ | ✓ | Cloud | Web | $6 |
| Elicit | Research & data agents | ✗ | ✓ | Cloud | Web | $49 |
| Factory | Coding agents | ✗ | ✓ | Both | CLI | $20 |
| GC AI | Research & data agents | ✗ | ✓ | Cloud | Web | $500 |
| GitHub Copilot | Coding agents | ✗ | ✗ | Cloud | IDE | $10 |
| GPT Researcher | Research & data agents | ✓ | ✓ | Self-host | Library | Free |
| Intercom Fin | Support agents | ✗ | ✓ | Cloud | Web | $0.99 |
| Juggler | Coding agents | ✓ | ✗ | Self-host | Visual | Free |
| LangChain | Agent frameworks | ✓ | ✓ | Self-host | Library | $39 |
| Letta | Agent frameworks | ✓ | ✓ | Both | Library | $20 |
| Lindy | Agent platforms | ✗ | ✗ | Cloud | Visual | $49.99 |
| LiveKit Agents | Voice agents | ✓ | ✓ | Self-hosted or Cloud | API | $50 |
| LlamaIndex | Agent frameworks | ✓ | ✓ | Both | Library | $50 |
| Manus | Research & data agents | ✗ | ✓ | Cloud | Web | From $20 |
| Mastra | Agent frameworks | ✓ | ✓ | Both | Library | $250 |
| Microsoft AutoGen | Agent frameworks | ✓ | ✗ | Self-host | Library | Free |
| n8n | Agent platforms | ✓ | ✓ | Both | Visual | €20 |
| OpenAI Agents SDK | Agent frameworks | ✓ | ✗ | Self-host | Library | Free |
| OpenAI Codex CLI | Coding agents | ✓ | ✓ | Local + Cloud | CLI | $8 |
| OpenCode | Coding agents | ✓ | ✓ | Local | CLI | $5 |
| OpenHands | Coding agents | ✓ | ✓ | Both | Web | Free |
| Perplexity | Research & data agents | ✗ | ✓ | Cloud | Web | $20 |
| PolyAI | Voice agents | ✗ | ✓ | Cloud | Web | Free |
| Pydantic AI | Agent frameworks | ✓ | ✗ | Self-host | Library | Free |
| Qualified | Sales & marketing agents | ✗ | ✓ | Cloud | Web | Custom |
| Relevance AI | Agent platforms | ✗ | ✓ | Cloud | Visual | From $19 |
| Replit Agent | Coding agents | ✗ | ✗ | Cloud | Web | $20 |
| Retell AI | Voice agents | ✗ | ✓ | Cloud | Web | $0.07–$0.31 |
| Rox | Sales & marketing agents | ✗ | ✗ | Cloud | Web | $50 |
| Sierra | Support agents, Agent platforms | ✗ | ✗ | Cloud | Visual | Custom |
| Unify | Sales & marketing agents | ✗ | ✗ | Cloud | Web | $20 |
| Vapi | Voice agents | ✗ | ✓ | Cloud | API | $0.05 |
| Windsurf (now Devin Desktop) | Coding agents | ✗ | ✗ | Both | IDE | $20 |
| Zot | Coding agents | ✓ | ✓ | Local | CLI | Free |
Methodology
This report is computed directly from The Agents Index’s quality-gated listings corpus (tools verified 2026-07-23–2026-07-24, per tool — see each listing’s own verification date), by the generator at scripts/generate-agents-report.ts:
- Open source / deployment / interface / pricing model are read from each listing’s structured
attributesdatasheet block — never inferred from prose. - API availability is marked only where the listing’s structured
apiAvailablefield is set. - Capability tags (Finding 6) are counted only from the structured
tagSlugsfield, kept separate from theattributesstats above since the two are independently-assigned dimensions. - Entry price in the full spec table is each tool’s own lowest real paid tier
priceDisplaystring, shown verbatim — this report does not convert per-token/per-minute/per-resolution/Custom pricing into a shared monthly unit, because doing so would misrepresent tools that don’t actually price that way.
For the full, sourced entry for each tool — including pros, cons, verdicts, and FAQ — see The Agents Index’s directory.
See Also
- The Agents Index: Full Directory — the complete, quality-gated directory of 51 AI agents and agentic tools with detailed breakdowns, verdicts, and sourced facts.
- Terminal AI coding agents: open, portable, or locked in? — our companion piece on why open-source license, model portability, and self-hosting are three separate questions.
- How The Agents Index researches AI agents — our sourcing and verification standard for every listing behind these numbers.