Maxim AI
The only eval platform we have scored that also ships a serious open-source gateway — Bifrost, Apache-2.0, 7,300 GitHub stars, 1,000+ models and an MCP gateway — wrapped around per-seat billing and three-day log retention on the free tier, so the bill tracks your headcount rather than your traffic.
PRICING
| Developer | Free | Up to 3 seats, 1 workspace, 10k logs/month, 3-day retention, no log overages allowed |
| Professional | $29/seat/mo | Unlimited seats, 3 workspaces, 100k logs/month, 7-day retention, simulation runs, online evals; 14-day trial |
| Business | $49/seat/mo | Unlimited workspaces, 500k logs/month, 30-day retention, RBAC, PII management, scheduled runs, custom dashboards |
| Enterprise | Custom | In-VPC deployment, SAML SSO, custom retention and log limits, audit logs, SOC 2 Type II / ISO 27001 / HIPAA / GDPR, custom BAAs, dedicated CSM |
Billing is per seat, not per unit of traffic, which is the opposite shape to most observability tools — a 20-person AI team on Business pays roughly €900/month before a single overage. Log overages are $1 per 10k logs on paid tiers; the free tier blocks overages entirely. Self-hosting is Enterprise-only, at undisclosed price.
checked 2026-08-16 · vendor pricing page
Element scores
Strengths
One loop instead of three tools: a versioned prompt IDE, agent simulation across persona and scenario libraries, and production tracing with sessions, spans and 1MB trace elements — with the same evaluators running in CI, in staging simulations and against live traffic. Evaluators come pre-built or custom across LLM-as-judge, statistical, programmatic and human, and are versioned so judges can be realigned to human preference as the agent drifts. SDKs cover Python, TypeScript, Java and Go, and Bifrost — the Apache-2.0 gateway with 7,319 GitHub stars — adds 23+ providers, automatic failover, per-team budgets and a genuine MCP gateway for free. Compliance is verified rather than claimed: SOC 2, ISO 27001, HIPAA and GDPR are all listed on a public trust center.
Honest dings
Per-seat pricing means cost scales with headcount rather than usage, which is backwards for an observability tool and punishes exactly the cross-functional collaboration Maxim markets. Log retention is 3 days free, 7 days at €27/seat and 30 days at €45/seat; anything longer, plus SSO, in-VPC and every compliance certificate, is Enterprise-only. Simulation runs and online evals — the differentiators — are paywalled above the free tier. Nothing but Bifrost is open source, so self-hosting the platform means an enterprise contract. Public review volume is thin: 4.8 on G2 from three reviews, 3.7 on Trustpilot, and G2's aggregated con is documentation quality.
Sources (11) — every claim traceable
Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).
- getmaxim.ai/pricing — Live pricing 16/08/2026: Developer free (3 seats, 10k logs, 3-day retention), Professional $29/seat, Business $49/seat, Enterprise custom; retention, RBAC, PII, SSO and in-VPC gating per tier (accessed 2026-08-16)
- getmaxim.ai/docs/introduction/overview — Official platform overview: Playground++ experimentation, evaluator framework, observability repositories and distributed tracing, and the data engine for synthetic and curated datasets (accessed 2026-08-16)
- getmaxim.ai/products/agent-observability — Official: sessions/traces/spans model, 1MB trace elements, OpenTelemetry ingest and forwarding to New Relic, online evaluation sampling, Slack and PagerDuty alerting (accessed 2026-08-16)
- getmaxim.ai/products/agent-simulation-evaluation — Official: AI-powered multi-turn simulation across personas, pre-built and custom evaluators (AI, human, programmatic), evaluator versioning, dataset curation and synthetic generation, CI/CD via GitHub Actions, Jenkins and CircleCI, SDKs in Python, TypeScript, Java and Go (accessed 2026-08-16)
- github.com/maximhq/bifrost — Bifrost repository: Apache-2.0, 7,319 stars, 23+ providers and 1,000+ models behind one OpenAI-compatible API, mcp-client / mcp-gateway / mcp-server topics, semantic caching, npx or Docker start (accessed 2026-08-16)
- trust.getmaxim.ai — Official trust center: legal entity H3 Labs Inc., SOC 2 / ISO 27001 / HIPAA / GDPR listed compliant, named customers including EY, Mindtickle, Atomicwork, Clinc and Qure (accessed 2026-08-16)
- elevationcapital.com/portfolio/maxim — Investor page: $3M seed led by Elevation Capital, founded 2023 by Vaibhavi Gangwar and Akshay Deo (Google and Postman), offices in India and the US (accessed 2026-08-16)
- economictimes.indiatimes.com/tech/funding/ai-app-ev… — Independent reporting of the seed round (18/06/2024), San Francisco base, and the mid-market and enterprise focus across SaaS, BFSI, healthcare and edtech (accessed 2026-08-16)
- g2.com/categories/ai-agent-observability — Independent review aggregation: Maxim AI 4.8/5 from only three reviews, listed pros ease of use, integrations and alerting, listed con poor documentation (accessed 2026-08-16)
- paperclipped.de/en/blog/ai-agent-evaluation-tools-c… — Independent comparison (22/03/2026) against Langfuse, Braintrust, Arize Phoenix and Confident AI: confirms Maxim is not open source, self-hosting is enterprise-only, and agent trace capture runs via the Bifrost gateway (accessed 2026-08-16)
- trustpilot.com/review/getmaxim.ai — Independent review platform: 3.7 rating on thin volume, including a March 2026 reviewer who switched from Braintrust over missing functionality (accessed 2026-08-16)