Opper
Overview
Opper is a Stockholm-based unified AI gateway and agent control plane. Founded in 2023 by Johan Gustafsson and Göran Sandahl (previously co-founded Unomaly, acquired by LogicMonitor in 2020), Opper routes requests across 301 models from 21 providers through a multi-API-compatible gateway (OpenAI, Anthropic, and Google API formats). Beyond basic gateway routing, Opper differentiates with a "control plane" layer offering observability, in-context learning, guardrails, and agent SDKs.
Raised €1.6M pre-seed (April 2024) and $3M oversubscribed pre-seed (July 2025) from Luminar Ventures, Emblem Venture Capital, Greens Capital, and angel investors (including backers of Lovable).
Markets
- Primary: Developers and engineering teams building AI-powered applications in production
- Segments: Startups needing multi-model access, enterprises requiring compliance/guardrails, agent builders
- Geographic focus: EU-first (AWS Stockholm hosting, EU data residency), expanding globally
- Verticals: General-purpose but emphasizes regulated industries needing PII masking, audit trails, and data sovereignty
Products
Unified AI Gateway
- 301 models across 21 providers (OpenAI, Anthropic, Google, Mistral, xAI, Fireworks, Cerebras, Groq, Novita, Evroc, Azure, AWS, Nebius, Arcee, Berget, Perplexity, Pruna, PersonaPlex, Docling, ElevenLabs, Inceptron)
- Drop-in OpenAI SDK compatibility
- Anthropic Messages API compatibility (V3 API)
- OpenAI Responses API compatibility (V3 API)
- Google Interactions API compatibility (V3 API)
- Roundtable API (V3 beta) — ensemble multi-model calls with resolution strategies (summary, fast, multiple_choice)
- Realtime WebSocket for voice agents
- Automated fallbacks and retries
- Model aliases with failover
- Streaming support including structured JSON streaming
- BYOK (Bring Your Own Key) support
Control Plane
- Observe — Call tracing, session scoring, quality metrics, cost tracking
- Steer — In-context learning from examples, feedback loops, prompt optimization
- Guard — PII masking, content filtering, enterprise guardrails (available via API headers; dedicated UI coming soon)
- Comply — Budget caps, audit trails, compliance tools
- Route — Intelligent model routing, switch models without code changes
- Memory — Custom knowledge storage and retrieval (RAG-like)
Agent SDKs
- Python and TypeScript agent frameworks
- MCP integration support
- Multi-agent composition and delegation
- Built-in observability for agent traces
Opperator
- Terminal-first framework for building and operating AI agents
- Local execution with built-in daemon system
- Template-based code generation, secret management, integrated debugging
- Managed infrastructure for agent deployment
IDE & Tool Integrations
- OpenCode, Continue.dev, Cursor, Cline, pi, OpenClaw, Hermes, Vercel AI SDK support
- Skills framework compatible with Claude Code, Cursor, Cline, GitHub Copilot, Windsurf, OpenAI Codex
Playground
- Side-by-side model comparison tool
Supported Models
| Provider | Count | Notable Models |
|---|---|---|
| 61 | Gemini 2.5, 3 Flash/Pro Preview, 3.1 Flash/Pro Preview, Imagen 4, Gemma 4 (26B MOE + 31B), Claude via GCP/Vertex | |
| OpenAI | 52 | GPT-5/5.1/5.2/5.3/5.4, GPT-4o, o1, o3, o4-mini, Sora 2, Whisper, TTS |
| xAI | 24 | Grok 3/4/4.1/4.20 variants, Grok Code Fast 1, Grok Imagine (image/video), TTS |
| AWS | 21 | Claude (incl. Opus 4.7), Mistral, Nemotron, GPT-OSS via Bedrock |
| Azure | 16 | GPT-5, GPT-5.1, GPT-5.1 Codex Mini, DALL-E 3, Claude via Azure-Anthropic |
| Nebius | 27 | DeepSeek V3.2, GLM 5, Hermes 4, INTELLECT-3, Kimi K2.5, MiniMax M2.5, Nemotron 3, Qwen 3/3.5 Next |
| Novita | 17 | DeepSeek, GLM, Kimi K2.5/K2.6, MiniMax, Qwen3 Coder/3.5 |
| Mistral | 12 | Mistral Large 3, Small 4, Codestral, Magistral |
| Evroc | 11 | EU-hosted: GPT-OSS, Kimi K2.5, Phi-4, Qwen3 |
| Fireworks | 16 | DeepSeek V3.2, GLM-5, GLM 5.1, GPT-OSS 120B/20B, Kimi K2.5/K2.6, MiniMax M2.5/M2.7, Qwen3-8B, Qwen3-VL-30B, Qwen 3.6 Plus |
| Anthropic | 9 | Claude Opus 4.7, Opus 4.6, Sonnet 4.6, Opus 4.5 |
| Groq | 8 | GPT-OSS, Llama, Kimi K2, fast inference |
| Berget | 5 | EU-hosted: GLM-4.7, GPT-OSS, Llama, Mistral |
| Cerebras | 4 | GLM-4.7, GPT-OSS, Qwen3 |
| Perplexity | 4 | Sonar, Sonar Pro, Deep Research |
| Pruna | 4 | P-Image, VACE, Wan video generation |
| Arcee | 3 | Trinity Large Preview/Thinking, Mini |
| Docling | 2 | Document processing (OCR) |
| ElevenLabs | 1 | Conversational AI (audio) |
| Inceptron | 3 | GLM 5.1 FP8, Llama 3.3 70B Instruct, MiniMax M2.5 |
| PersonaPlex | 1 | Voice model |
Total: 301 models across 21 logical providers (25 API-level)
Last verified: 2026-04-21
Key Capabilities
| Capability | Status | Notes |
|---|---|---|
| Multi-model routing | ✅ | 301 models, 21 providers |
| Ensemble/Roundtable | ✅ | V3 beta — parallel multi-model calls with consensus strategies |
| OpenAI compatibility | ✅ | Drop-in SDK replacement (Chat Completions + Responses API) |
| Anthropic compatibility | ✅ | Messages API via V3 compat layer |
| Google compatibility | ✅ | Interactions API via V3 compat layer |
| Realtime voice | ✅ | WebSocket endpoint for voice agents |
| Streaming | ✅ | Including structured JSON streaming |
| Fallbacks/retries | ✅ | Automated with model aliases |
| Observability/tracing | ✅ | Full call tracing, cost tracking |
| In-context learning | ✅ | Functions learn from examples |
| PII masking | ✅ | Via API headers (Guard control plane UI still coming soon) |
| Budget controls | ✅ | Strict caps and audit trails |
| Agent SDKs | ✅ | Python + TypeScript with MCP |
| BYOK | ✅ | Use own provider keys |
| Embeddings | ✅ | Via API |
| Image generation | ✅ | DALL-E 3, Imagen 4 |
| OCR | ✅ | DeepseekOCR |
| EU data residency | ✅ | AWS Stockholm default |
| SSO/SAML | ✅ | Enterprise tier |
Last verified: 2026-04-17
Pricing
| Tier | Price | Notes |
|---|---|---|
| Starter | Base model cost + 3% fee | 200+ models, 30-day retention, community support |
| Utility | Base model cost + 3% fee + per-request control plane fees | All Starter + Observe/Route/Steer/Guard/Comply features, email support |
| Enterprise | Custom | Custom retention, SLA, dedicated support, SSO/SAML, custom regions, volume discounts |
Last verified: 2026-03-25
URLs to Monitor
| URL | Label | Notes |
|---|---|---|
https://docs.opper.ai/sitemap.xml |
Sitemap | Docs sitemap for page discovery |
https://opper.ai/changelog |
Changelog | Product updates and releases |
https://opper.ai/pricing |
Pricing | Pricing tiers and fees |
https://docs.opper.ai/control-plane/observe |
Observe | Observability features |
https://docs.opper.ai/control-plane/guard |
Guard | Guardrails and safety |
https://docs.opper.ai/agents/overview |
Agent SDK | Agent framework docs |
https://docs.opper.ai/api-reference |
API Reference | API endpoints |
https://docs.opper.ai/capabilities/models |
Supported Models | Dynamic model catalog (301 models) |
https://docs.opper.ai/control-plane/route |
Route | Intelligent routing (coming soon) |
https://docs.opper.ai/control-plane/steer |
Steer | In-context learning feature |
https://docs.opper.ai/opperator/overview |
Opperator | Terminal agent framework |
https://docs.opper.ai/overview/integrations |
Integrations | IDE and tool integrations |
https://docs.opper.ai/v3-api-reference/roundtable/create-roundtable |
Roundtable API | Ensemble multi-model call endpoint |
https://docs.opper.ai/v3-api-reference/compatibility/openresponses |
OpenResponses API | OpenAI Responses API compatibility endpoint |
Strategy
- EU-first positioning: Differentiates on data sovereignty with Stockholm-based hosting, zero data retention options, and EU inference providers (Evroc). Appeals to European enterprises with strict GDPR requirements.
- Control plane upsell: Free/cheap gateway layer (3% markup) draws users in; monetizes via per-request control plane features (observe, steer, guard, comply). This is a unique angle vs pure gateways.
- Agent ecosystem play: Investing heavily in agent SDKs (Python + TypeScript) with MCP integration, positioning as infrastructure for the emerging AI agent market.
- In-context learning: Unique "Steer" feature lets functions improve from feedback/examples without retraining — differentiator vs static prompt routing.
- Coding agent integrations: "Opper Skills" bring platform capabilities directly into coding agents (Cursor, Windsurf, etc.)
- Cost reduction narrative: Blog content emphasizes 98.6% cost reduction using their routing + smaller model optimization.
Formidability
Score: 4/10
Opper has interesting differentiation (control plane, in-context learning, EU-first) but is very early stage ($4.6M total raised, small team). The 290+ model count and 3% markup are competitive with OpenRouter's model on paper, but Opper's traction and brand recognition are far behind. The control plane features (observe, steer, guard) add unique value that pure gateways don't offer, but the market hasn't validated demand for this bundled approach yet. Watch for: enterprise adoption traction, model count growth, and whether the control plane features gain developer mindshare.