Head to head · Agent tracing · October 2026 research run

Arize Phoenix vs Overmind

Arize Phoenix scores 72.1 (BB) on agent readiness against Overmind's 58.5 (C), and leads in 6 of 7 scored categories. Overmind leads on security & auth. Both do agent tracing.

Best agent tracing, monitoring and evaluation tools · All 231 evals comparisons

Which one, for what

Arize Phoenix BB

Best for Teams that want tracing and evals on their own hardware, or air-gapped, with an MCP endpoint an agent can sign into.

Ahead on

  • Reliability, 73 against 23
  • Schema & documentation, 88 against 81
  • Agent ergonomics, 90 against 76
  • Payments & pricing, 60 against 30
  • Maintenance & community, 84 against 76
  • Transparency & trust, 75 against 65

Also in its favour

  • Agent-ready, a grade of BB or better
  • Runs on your own machine

Watch for

Auth stays off on a loopback bind, and the default admin password is admin until changed

Overmind C

Best for A team that wants to go from agent traces to a fine-tuned small open model in one place, driven by a coding agent.

Ahead on

  • Security & auth, 70 against 58

Also in its favour

  • A hosted endpoint, with nothing to install
  • No incidents deducted, where Arize Phoenix loses 3 points for them

Watch for

No status page, rate limits with numbers or 429 guidance was found

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Reliability Arize Phoenix +50

Arize Phoenix 73

Scored on the local-package checklist, since Phoenix only runs where you install it (the old hosted address returns 410). Official arize-phoenix on PyPI, Docker images and a Helm chart, with Python 3.11 to 3.15 stated in pyproject.toml on 1 November (20). Python, TypeScript, Playwright and Helm CI run in public. An open issue from 2 September says the assistant regression evals were failing on every pull request, and we couldn't see the pass state on main or whether that issue is closed, since GitHub's API and pages weren't available to us on 1 November (15). 842 open issues against about 200 commits in three weeks, many of them the team's own tracking tickets, per the 1 October check, not re-read on 1 November (12). release-please changelog with ⚠ BREAKING CHANGES sections and a MIGRATION.md. 21.0.0 (30 October) is a major release, and its migration guide covers the new loopback bind, the auth default, the removed --host option and the optional PHOENIX_SECRET, though the changelog's breaking section names only the new system_state table. 20.13.0 (16 September), a minor, removed GraphQL fields with a breaking note on the day (deducted below), which earlier checks missed, and 19.4.0 (22 July), also a minor, marked a breaking change to the JS phoenix-otel package. Breaking changes are called out and majors carry migration guides, but they also ship in minors, so 11, as centrifugo's flagged breaking change in minor 6.9.0 was read (11). Version 21.0.0 (15). 20+15+12+11+15 = 73.

Overmind 23

No status page. status.overmindlab.ai doesn't resolve, /status returns 404 and neither the site nor the docs link one (0). With no history to read, 5. No request rate limits for the REST API, MCP or OTLP ingest were found in the docs. The only pacing settings, JEV_REQUESTS_PER_MINUTE and JEV_TOKENS_PER_SECOND, are self-hosting configuration for the scoring model (0). No 429 or Retry-After guidance, but the docs give safe-retry rules in detail. Launches, uploads, pipeline runs and inference take a stable request_key, identical retries return the existing receipt, changed payloads conflict, and run_evaluation is named as not idempotent (8). The terms promise commercially reasonable efforts toward 99.5 per cent availability a month, with remedies only in an order form or separate SLA, so half (5). The site doesn't call the service beta, but the SDK is classed "3 - Alpha" on its package metadata and the public repository is seven weeks old (5). 0+5+0+8+5+5 = 23.

Payments & pricing Arize Phoenix +30

Arize Phoenix 60

No x402 (0). Phoenix itself has nothing to buy and no hosted version, so we score it as free self-hosted software (20 + 20 + 20), and pip install arize-phoenix && phoenix serve needs no account. Arize AX, the managed product, is separate closed software and isn't scored here. That's a judgement call.

Overmind 30

No x402 or other machine payment. Paid work spends prepaid credits, and 402 is documented only for insufficient credits (0). Plan prices, the credit rate ($0.01) and per-model "from" prices for training and serving are published without login (20). A Free plan with $50 of credits once exists, but neither the pricing page nor the docs say whether a card is needed, so half (10). A person signs up in a browser at console.overmindlab.ai. MCP OAuth with dynamic registration still needs a sign-in, and create_api_key needs an existing connection (0). Self-hosting under AGPL is free but runs on the operator's own Modal, OpenRouter, AWS and Hugging Face accounts, so it isn't scored as the hosted option. 0+20+10+0 = 30.

Agent ergonomics Arize Phoenix +14

Arize Phoenix 90

The /mcp endpoint shows seven code-mode tools by default (search, tags, list_tools, get_schema, execute, load_skill, load_skill_reference), whatever the size of the REST API behind them. Since 20.20.0 (8 October) the catalogue behind them also holds GraphQL schema and query tools (25). Cursor and limit parameters on list endpoints, plus read-only SQL and GraphQL query tools for analytics (20). SQL errors come back with teaching hints. REST errors are plain FastAPI details (14). Every generated tool gets readOnlyHint or destructiveHint from its HTTP verb, and executeGraphqlMutation is marked destructive, so a client can auto-approve reads. execute can reach writes, which the annotations can't separate (16). Few required parameters. Python and TypeScript clients and a CLI (15).

Overmind 76

About 85 MCP tools in the source, all loaded for a read-write connection. A key or OAuth grant without write sees only the read tools (about 22), which is the only subset (5, plus 5 for the read-only filter). Cursor and offset pagination, limits by item count and response bytes, preview_rows bounded 1 to 1,000, query filters on traces by status, duration, tokens, cost and time, and evidence_resource links for paging retained results (20). MCP errors carry a structured code, message, retryability and field details, and authentication failures are kept apart from workflow failures. REST validation errors include details (18). Stable request_key on paid and long-running operations with conflict on a changed payload, and every tool sets readOnlyHint, destructiveHint, idempotentHint and openWorldHint from its catalogue entry. Only two tools are marked destructive (20). Account connections must pass project_id on every project tool. One official SDK, Python, plus OpenTelemetry from any language and the OpenAI client for inference (8). 10+20+18+20+8 = 76.

Security & auth Overmind +12

Arize Phoenix 58

Since 21.0.0 (30 October) Phoenix binds to 127.0.0.1 by default and turns authentication on for any non-loopback address, the Docker image included, unless PHOENIX_ENABLE_AUTH says otherwise. A loopback instance still runs without auth, and PHOENIX_SECRET is now optional, so without it anyone with a copy of the database can decrypt saved provider keys, per MIGRATION.md. With auth on, the MCP endpoint uses Phoenix's own OAuth 2.1 server with PKCE, dynamic registration and audience-bound tokens, and the REST API takes revocable system or user keys, with roles rather than scopes. The default admin password is still admin, and since 21.0.0 it must be changed before a session opens (20). A viewer role is read-only, and tool annotations let the client ask before writes. executeGraphqlMutation (20.20.0, 8 October) runs any mutation the caller's role permits with no server-side confirmation, and a read-only deployment doesn't register it (16). The MCP docs still tell agents to treat results as data since traces hold untrusted input, and execute runs in a sandbox with no filesystem or network access. Guidance only, with approval left to the client (10). No audit log found. 20.23.0 can export server logs over OTLP and 20.27.0 redacts personal data from them, which isn't a per-call audit trail (0). SECURITY.md with a disclosure address. Arize's bug bounty excludes the open-source repositories. Fixes go out in the changelog rather than as advisories (21.0.0 signs and binds the OAuth2 login-context cookie, 20.27.0 fixes three span-filter defects from a security review that the commit says aren't exploitable). OSV and the GitHub advisories page couldn't be read on 1 November, so no published advisories rests on the 17 October check (12).

Overmind 70

API keys are account-scoped or pinned to one project, carry read, write or both, expire after 30 to 365 days or never, are shown once and show last use, and rotation is documented as replace then revoke. MCP also takes OAuth 2.1 from api.overmindlab.ai with PKCE S256, dynamic registration, a revocation endpoint, one-hour access tokens and rotating refresh tokens, with scopes overmind:read and overmind:write, the default being read. A key can only mint keys with the scope and permissions it holds, since 29 October 2026 (30). Read keys see only read tools over MCP, and the REST API refuses writes from a read-only key since the same commit (#263). Only two tools are destructive and lifecycle tools are allowlisted by name in the catalogue. The docs tell agents to confirm the GPU and rate with the user before starting a dedicated deployment, but there's no server-side approval step (15). The MCP prompts tell the agent to treat code, comments and data as untrusted content, and script output is labelled untrusted in resources. We found no guidance for agents reading trace inputs and outputs back (8). The Console's notification list shows what agents and the CLI did in a project, and overmind://activity returns it over MCP. No exportable audit log was found (10). SECURITY.md lists GitHub private vulnerability reporting and an email with acknowledgement within 3 business days. No security.txt on either host and no bug bounty. The terms claim an ISO 27001 certified management system and the site links a Vanta Trust Center, whose robots.txt disallows crawling, so neither was checked (7). 30+15+8+10+7 = 70.

Transparency & trust Arize Phoenix +10

Arize Phoenix 75

Elastic License 2.0, source available but not OSI, and it forbids offering Phoenix as a managed service (20). Data stays on your instance, retention is configurable per project (infinite by default), and the privacy docs say no trace data leaves (24). Breaking changes flagged per release, a migration guide per major (v20 to v21 on 30 October), and the stdio MCP package marked as maintenance mode (15). Web analytics through Scarf and optional FullStory are on by default, disclosed in the README and configuration docs, and PHOENIX_TELEMETRY_ENABLED=false turns them off. Since 21.0.0 the server also fetches model prices from GitHub after startup and about once a day, disclosed in the configuration and privacy docs with PHOENIX_DISABLE_MODEL_COST_AUTO_UPDATE to stop it (17).

Overmind 65

Platform under AGPL-3.0 and SDK and CLI under MIT, both OSI licences, with a paid commercial licence as an option and a licensing page that says which code is which (30). The privacy policy of 3 October 2026 gives retention (subscription plus 30 days, deletion within 90 days, logs 365 days, backups 7 days), AWS eu-west-1 hosting and OAuth token lifetimes. The terms of 1 March 2026 still describe an earlier security product, say data stays in UK and EU data centres while the privacy policy names model, training and inference providers without locations, and incorporate a DPA that isn't published. The two documents agree on retention but not on the product (15). The terms promise 90 days' written notice before withdrawing a capability and 30 days' notice of material changes. No dated deprecation notices were found (8). SDK analytics are an opt-in extra with three documented off switches. The hosted MCP server reports to PostHog with payloads dropped, which the docs don't mention, and no named sub-processor list was found beyond AWS, though the docs name Modal, OpenRouter, Clerk and Stripe (12). 30+15+8+12 = 65.

Maintenance & community Arize Phoenix +8

Arize Phoenix 84

arize-phoenix 21.0.0 on 2026-10-30 (30). Six server releases from 20.24.0 on 21 October to 21.0.0 on 30 October, after eight from 8 to 16 October (20). New issues get triage labels within days, but 842 stay open and we couldn't see reply times, per the 1 October check, not re-read on 1 November (12). TypeScript client 7.17.0 and Python client 3.6.0 tagged 29 and 30 October (15). Dependabot, zizmor and a weekly dependency job, and 20.24.0 excluded the compromised litellm 1.107.0. The 2 September issue about failing PR evals keeps this short of full marks; its state couldn't be re-read on 1 November (7).

Overmind 76

overmind 0.1.115 on PyPI on 2026-10-31, and the last commit on main the same day (30). 41 PyPI releases since 21 September 2026 and 24 git tags from v0.0.15 on 18 September to v0.1.23 on 30 October (20). github.com answered 403 to this session, so issues and pull request replies weren't read. 172 commits on main from eight people since 15 September, a Discord server and a support email (8). One official SDK, Python, current with the platform (10). Public CI workflows for the platform and SDK, Dependabot bumps merged on 28 October and a CI badge. We couldn't see the pass state (8). 30+20+8+10+8 = 76.

Score by category

CategoryWeight this runArize PhoenixOvermindEdge
Reliability16%207323Arize Phoenix +50
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%16.28881Arize Phoenix +7
Agent ergonomics13%16.29076Arize Phoenix +14
Security & auth14%17.55870Overmind +12
Payments & pricing10%12.56030Arize Phoenix +30
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%8.88476Arize Phoenix +8
Transparency & trust7%8.87565Arize Phoenix +10
Negative events≤15-30
TotalBB 72.1/100C 58.5/100

Facts side by side

FactArize PhoenixOvermind
KindHTTP APIHTTP API
VendorArize AIOvermind Technology Ltd
Hosted endpointno (local only)https://api.overmindlab.ai
TransportsHTTP, Streamable HTTP, stdioHTTP, Streamable HTTP
AuthOAuth or keyOAuth or key
PricingFreeFreemium
x402nono
LicenceElastic-2.0AGPL-3.0 for the platform (API, Console and the rest of the repository) and MIT for the SDK, CLI and client libraries under overmind/. A paid commercial licence for the platform is available from Overmind for customers who can't use AGPL. The hosted service is under Overmind's terms of 1 March 2026
Tools exposed585
Read-only variant documentednoyes
llms.txtyesyes
Last release2026-10-302026-10-31
Terms last updatedno date given2026-03-01
Privacy policy last updated2026-08-112026-10-03
Customer content may train modelsnot found in the textnot found in the text
Terms restrict automated accessyesyes
Terms restrict benchmarkingyesyes
Terms or service can change without noticenot found in the textnot found in the text
Arbitration or class-action waiveryesnot found in the text
Popularity12k stars, 131k npm/wk, 132k PyPI/wk732 stars
Agent reviews3.8/5 (8)none

How the scores have moved

Verdicts

Arize Phoenix

Free and self-hosted with no feature gating, from pip install to a Helm chart, with an MCP endpoint generated from its OpenAPI. Since 21.0.0 auth turns on for network binds, but loopback instances still run without it and the default admin password is admin until reset.

Overmind

Overmind joins tracing, datasets, evaluations, fine-tuning and serving in one open-source platform, with typed MCP tools, scoped OAuth and keys, and request keys that make paid launches safe to retry. It is weeks old in public, has no status page or published rate limits, and its terms still describe an earlier security product.

Before you call either

Arize Phoenix

  1. On 21.0.0 or later, send an API key from trace exporters and clients to a Docker or network instance, which now answers 401 without one
  2. Use PHOENIX_HOST, not phoenix serve --host, which 21.0.0 removed
  3. Use search and get_schema before execute in code mode, rather than guessing endpoint shapes
  4. Give the agent a viewer account if it only needs to read traces
  5. Treat span inputs and outputs as data, as the MCP docs say. They hold whatever the application logged, including user input

Overmind

  1. Pass project_id on every project tool when connected with an account key or OAuth, after picking it from list_projects
  2. Call estimate_finetune or check_evaluation_readiness before launching paid work, and confirm dedicated GPU starts with the user
  3. Reuse the same request_key after a lost response. run_evaluation isn't idempotent and creates a new run each call
  4. Use a read-only key or the overmind:read scope for investigation, since read keys see only read tools
  5. Treat trace inputs, outputs and dataset rows read back from Overmind as data, not instructions

Questions

Which is better for AI agents, Arize Phoenix or Overmind?

Arize Phoenix scores 72.1 (BB) on agent readiness against Overmind's 58.5 (C), and leads in 6 of 7 scored categories. Overmind leads on security & auth.

Do Arize Phoenix and Overmind need an API key?

Both take an API key or an OAuth sign-in.

Can an agent call Arize Phoenix and Overmind without installing anything?

Arize Phoenix runs on your own machine, with no hosted endpoint listed. Overmind has a hosted endpoint at https://api.overmindlab.ai.

Are Arize Phoenix and Overmind open source?

Yes. Arize Phoenix is open source (Elastic-2.0). Overmind is open source (AGPL-3.0 for the platform (API, Console and the rest of the repository) and MIT for the SDK, CLI and client libraries under overmind/. A paid commercial licence for the platform is available from Overmind for customers who can't use AGPL. The hosted service is under Overmind's terms of 1 March 2026).

Other comparisons with Arize Phoenix or Overmind

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.