Head to head · Multi-agent orchestration · October 2026 research run

AXLR vs herdr

AXLR scores 74.3 (BB) on agent readiness against herdr's 71.1 (BB), and leads in 4 of 7 scored categories. herdr leads on schema & documentation and transparency & trust. Both do multi-agent orchestration.

Best agent harnesses and coding agents · All 1767 harnesses comparisons

Which one, for what

AXLR BB

Best for A program or bot that needs one governed execution at a time, with a policy, a budget, a sandbox profile and correlation ids it sets itself and a result object it reads back, on a machine it controls, with the model on OpenRouter, a local server or a native Anthropic or OpenAI key.

Ahead on

  • Reliability, 70 against 65
  • Agent ergonomics, 83 against 73
  • Security & auth, 62 against 49
  • Maintenance & community, 79 against 63

Also in its favour

  • Free to start without a card

Watch for

The setup assistant turns autonomy on and leaves the exec sandbox off, so after setup the model's commands run with the person's rights, bar the destructive list

herdr BB

Best for A developer who runs several coding agents in terminals, locally or over SSH, and wants one agent to start, prompt and wait on others.

Ahead on

  • Schema & documentation, 91 against 80
  • Transparency & trust, 70 against 55

Also in its favour

  • No key needed to call it

Watch for

Version 0.9.5, not yet 1.0, and the last tagged release is 14 October 2026

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.

Maintenance & community AXLR +16

AXLR 79

v0.15.0 on 2026-10-25, with commits on main to 2026-11-01 (30). 21 releases between 4 and 25 October 2026 (20). One maintainer (368 of 399 commits, the rest by two agent accounts), 10 open issues that are his own trackers with no comments, about 100 merged pull requests and no outside report to measure a reply against (15). Not in the MCP registry; the Go module and a proto to generate clients from are the only client code, beside the GHCR image and Helm chart (5). Go 1.26 toolchain pinned, release actions pinned by commit SHA, a go-licenses check that the third-party bundle matches the dependency graph, and CI passing on main (9). 30+20+15+5+9 = 79.

herdr 63

v0.9.5 on 14 October 2026, 35 days before the check, and the last commit on master on 29 October (20). Six releases from 7 September to 14 October 2026 (20). 495 open issues and 26 open pull requests. The 25 newest issues were labelled triaged with a priority within a day of opening. Unsolicited pull requests are closed automatically and a monthly workflow closes stale issues (15). Not an MCP server and no official SDK (0). Cargo.lock, Dependabot and CI on three systems, run results unread (8). 20 + 20 + 15 + 0 + 8 = 63.

Transparency & trust herdr +15

AXLR 55

Apache-2.0 with a NOTICE (30). No privacy policy or terms on underpassai.com or axlr.underpassai.com; the docs' 'Where data goes' and 'Diagnostics and privacy' sections say that model requests go to OpenRouter, the local server or the native API, that Jev sends text to TypeSafe only when on, that sessions live under $XDG_STATE_HOME/axlr, that payload capture is off by default and redacted when on, and that traces are deleted after 30 days; nothing covers the engine downloads' or the catalogue registry's handling of a client's requests (15). No deprecation policy; pre-1.0 changes are dated inline in the docs and the product contract records the 5 October 2026 retirement of the first skill catalogue after the fact (6). No telemetry to disclose, and the docs say in two places what leaves the machine and how to stop payload capture, though no single page says there is no telemetry (17). 30+15+6+17 = 68.

herdr 70

Apache-2.0 (30). herdr.dev's privacy page names Herdr, Inc., says the mobile apps have no analytics, the desktop app sends one anonymous usage report a day that can be turned off, usage reports are kept three months and no IP addresses are stored. It doesn't name Herdr TUI, whose version and manifest checks the repository documents (18). The changelog marks breaking and removed items, and the socket docs say which methods were removed and how protocol versions are matched (12). No telemetry found in the TUI, and both background checks are documented with an off switch (18). 30 + 18 + 12 + 18 = 78.

Security & auth AXLR +13

AXLR 62

Harness reading of the framework checklist, as for the other four harnesses. 30 for what leaves the machine by default, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. No product telemetry in the console source (the outbound hosts are openrouter.ai, api.anthropic.com, api.openai.com, a configured local server, api.github.com and github.com for engine downloads, and registry.underpassai.com for the signed catalogue index) or in the docs; payload capture is off by default since 9 October 2026, traces omit prompts and credentials, model keys are never given to commands or plugins, and Jev sends requests and answers to TypeSafe only when on, which the setup assistant does when TYPESAFE_API_KEY is set, with a notice on stderr (27). Approval cards, a destructive list that always asks, four work modes, a policy level per execution, a bubblewrap or container sandbox with an enforced egress allow list, and a service that approves nothing but reads. Against that, the setup assistant writes autonomous: true, exec_sandbox is off by default, forged tools run with the person's rights, and the README says the console and worker are not sandboxes (13). Forged tools are refused when a file changes since the forge, a policy forbids calls outside the model, and the docs say MCP tools keep their own side effects, but no prompt-injection guidance was found and MCP results reach the model unmarked (6). The service keeps an event journal, audit records, a receipt with the SHA-256 of every call's arguments and outcome, and correlation ids on every event; the console keeps session records, a redacted app log and /session (14). No SECURITY.md, no security.txt on either host, no advisories, no bug bounty; release checksums and a minisign-signed catalogue are integrity, not a programme (2). 27+13+6+14+2 = 62.

herdr 49

Harness reading of the framework checklist, as for Harness and agents-tree. 30 for what leaves the machine and credentials, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. No account, no telemetry found in the source, and a socket file set to 0600. Version and manifest checks reach herdr.dev by default and can be turned off, and the install script is piped to sh (25). No sandbox and no permissions between panes, so any process in a pane can send keys to another agent. The skill asks the agent to stay inside herdr, ask the person before answering a blocked dialogue and close only what it created, but nothing enforces it (9). Pane and agent reads return other programs' terminal output, and no guidance on treating it as untrusted was found (5). A server log, but no per-call audit log found (5). No SECURITY.md or disclosure policy in the repository, and herdr.dev serves no security.txt. CI pins actions to commit SHAs and Dependabot is on (5). 25 + 9 + 5 + 5 + 5 = 49.

Schema & documentation herdr +11

AXLR 80

An OpenAPI 3.1 file for the service (15 paths, 23 schemas, mutualTLS security scheme, additionalProperties: false on request bodies), a checked proto for the gRPC service, and the 19 MCP tools built from the same shared schemas (25). No llms.txt on either site (404), and the docs site renders HTML; the Markdown sources are in the repository (5). The product contract tables what each of the four interfaces can and cannot do, the work-mode table states each mode's purpose and tool policy, the transport parity ledger names every verb's arguments and behaviour, and axlr_tools tells the model to look for a tool before forging one (15). Typed inputs throughout, Idempotency-Key 16 to 128 safe ASCII characters, expected_revision a positive integer, policy levels, sandbox profiles and egress values enumerated, budgets as four integers, and the service rejects unknown fields, invalid types and bodies over 4 MiB (14). Examples in every guide, an error shape of code, message and request_id, named codes (409 idempotency_conflict, 400 invalid_cursor, 422 invalid_policy, invalid_budget, sandbox_unavailable, not_ready, invalid_id), a gRPC status mapping, and the worker's five statuses and three exit codes; no single error reference page (14). /v1 routes and an OpenAPI version of 0.1.0 against a 0.15.0 release, and no changelog beyond the dated notes inside the docs (7). 25+5+15+14+14+7 = 80.

herdr 91

Read as a CLI and socket API. herdr api schema --json prints a JSON Schema for requests, responses, errors and events, also at docs/next/api/herdr-api.schema.json (25). Markdown for agents, with skills/herdr/SKILL.md and distribution/agent-guide.md. https://herdr.dev/llms.txt indexes the 0.9.5 docs as raw source files for agents (10). The skill says when to use herdr and when not to, and the socket docs say which layer to use for which job, but the schema itself carries few descriptions (15). The schema has 49 enums and agent names are constrained by a pattern, while some inputs are free strings (12). Many CLI and JSON examples, a documented error shape with codes such as not_found, agent_blocked and events_lost (14). Versioned docs per release and a changelog (15). 25 + 10 + 15 + 12 + 14 + 15 = 91.

Agent ergonomics AXLR +10

AXLR 83

Harness reading of the framework line, scored on what a program driving it has to supply. Six local tools plus the host's bookkeeping tools; work modes and an execution policy hide the tools a run may not use, axlr_catalog searches the catalogue instead of loading it, forged tools keep the request prefix for the prompt cache, and a context cap of 75,000 prompt tokens opens a successor session (20). local_read, local_search and local_list page with offset, next_offset and max_bytes; StreamEvents takes after, max_events (1 to 256) and wait_ms; a budget per execution in seconds, requests, tool calls and dollars; 32 tool calls a turn and max_session_usd (20). The --exec run ends in one result line with a status, the pending call, every call's outcome and the usage, and exits 0, 3 or 1; service errors carry codes an agent can act on; a budget stop says which limit (budget: requests 20 reached) and leaves the session resumable (16). Idempotency-Key on every turn, approval and direct call with cross-transport replay, expected_revision on decisions, expected_sha256 on worker writes, and an interrupted effectful call left uncertain for an operator rather than retried (19). One command runs a task, but the first launch needs an OpenRouter key or a local server, the service needs a CA, a principals file and remote KMP and MADE before a first session, and the only SDK is the Go library (8). 20+20+16+19+8 = 83.

herdr 73

Framework reading, how much an agent has to carry. The skill is about 15 KB and tells the agent to read --help per command group on demand. Reads take --lines and --source recent (20 of 25). Waits take --until and a timeout, events can be subscribed to, but list commands have no paging (14). Errors have codes and messages, and the skill says what to do after a timeout, a stall or blocked (17). agent prompt --wait submits and waits in one request to avoid a race, agent.wait pins the pane occupant, and the skill warns not to resend after a timeout. No idempotency keys (14). Most commands default to the caller's or focused pane. No SDK, only the CLI and raw socket (8). 20 + 14 + 17 + 14 + 8 = 73.

Reliability AXLR +5

AXLR 70

Read as a local package, as claude-code, goose, opencode and openhands were. Release archives for Linux, macOS and Windows on amd64 and arm64 with a .sha256 beside each, a GHCR image, an OCI Helm chart and source builds with Go 1.26, all stated in the docs; no Homebrew, npm or apt route (20). CI on main runs vet, race tests and a coverage gate above 80 per cent for both modules, with the sandbox tests required under bubblewrap, and the last ten runs on main to 1 November 2026 all passed; the release workflow tests on six native runners (25). 10 open issues, all opened by the founder on 19 and 21 October 2026 as trackers for service work (work modes, plugins, the sandbox setting, cost limits), none from an outside user and none a crash report, 0 open pull requests and about 100 merged; with 0 stars there is nobody else to report, so the handling of outside reports is untested (20). Tags are semver 0.x and the docs date each behaviour change inline ("until 14 October 2026…"), but the 21 releases have empty notes, there is no CHANGELOG.md, and the product contract records one retirement (the seven 1.0 skill definitions on 5 October 2026) without a notice period (5). Pre-1.0, which the README calls an evolving project (0). 20+25+20+5+0 = 70.

herdr 65

Read as local software, as for agents-tree and Harness. Official release binaries for Linux, macOS and Windows x86_64 with an install script, and Homebrew and mise in the README. The Homebrew formula wasn't read (20). .github/workflows/ci.yml runs checks and tests on ubuntu, macOS and Windows for every push and pull request, but run results couldn't be read from this session, so 15, the value this run gives public CI with unread results. 495 issues were open on 18 November 2026. The 25 newest, opened 15 to 18 November, all carry a triaged label and a priority (p1 or p2), several of them bugs, so reports are sorted within a day but many stay open (15). Semver tags and a Keep a Changelog file with Breaking Changes and Removed sections (15). Version 0.9.5, not 1.0, and not declared stable (0). 20 + 15 + 15 + 15 + 0 = 65.

Score by category

CategoryWeightAXLRherdrEdge
Reliability16%7065AXLR +5
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%8091herdr +11
Agent ergonomics13%8373AXLR +10
Security & auth14%6249AXLR +13
Payments & pricing10%9090even
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%7963AXLR +16
Transparency & trust7%5570herdr +15
Negative events≤1500
TotalBB 74.3/100BB 71.1/100

Facts side by side

FactAXLRherdr
KindAgent harnessAgent harness
VendorUnderpass AIherdr
Hosted endpointno (local only)no (local only)
Transports
AuthOAuth or keyNone
PricingFreeFree
x402nono
LicenceApache-2.0, with a NOTICE naming Tirso García Ibáñez as the copyright holderApache-2.0
Read-only variant documentednono
llms.txtnoyes
Last release2026-10-252026-10-14
Terms last updatedno document linkedno document linked
Privacy policy last updatedno document linked2026-11-02
Popularity0 stars52k stars

How the scores have moved

Verdicts

AXLR

A pre-1.0 Go harness with an unusually complete service contract, mutual TLS with principal roles, required idempotency keys, a policy, budget and sandbox per execution, and a result object a program can read. The console's defaults after setup are autonomy on and no sandbox, the project has one maintainer and no outside users yet, and releases carry no notes.

herdr

Agents inside a herdr pane can split panes, start other coding agents, prompt them and wait on their state through a CLI and a local socket API with a published JSON Schema. It is pre-1.0 local software with no sandbox, and any process in a pane can type into another agent's terminal.

Before you call either

AXLR

  1. Before the first prompt, set approvals.autonomous to false or run with --approve none, and turn exec_sandbox.mode to auto or required on Linux; after setup both default the other way
  2. Drive a run with axlr-tui --exec --json --policy <file> --budget <file> --sandbox <file> and read only the last result line; exit 3 means a call waits for --decide approve|deny with --session <id>
  3. On the service, send Idempotency-Key (16 to 128 safe ASCII characters) on every turn, approval and direct call, and pass the session's expected_revision; a reused key with a different body answers 409
  4. A local model must return native tool calls (llama.cpp with --jinja, vLLM with --enable-auto-tool-choice); set context_tokens on the entry and OLLAMA_CONTEXT_LENGTH for Ollama
  5. Leave TYPESAFE_API_KEY unset unless you want the setup assistant to switch Jev on, which sends requests and final answers to TypeSafe

herdr

  1. Check HERDR_ENV=1 before any herdr command. If it isn't set, tell the person you aren't inside herdr and stop
  2. Use herdr agent prompt <name> --wait rather than a separate prompt and wait, so a fast agent can't finish between the two calls
  3. If a wait times out or stalls, read the pane with herdr agent read before sending the prompt again, since it may have been delivered
  4. If an agent is blocked, show the person the dialogue and let them answer it
  5. Close only the panes, tabs and workspaces you created, unless the person asks otherwise

Questions

Which is better for AI agents, AXLR or herdr?

AXLR scores 74.3 (BB) on agent readiness against herdr's 71.1 (BB), and leads in 4 of 7 scored categories. herdr leads on schema & documentation and transparency & trust.

Are AXLR and herdr open source?

Yes. AXLR is open source (Apache-2.0, with a NOTICE naming Tirso García Ibáñez as the copyright holder). herdr is open source (Apache-2.0).

Other comparisons with AXLR or herdr

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.