Hatchet runs your tasks fast. Avalanche runs the work your agents do.

Quick answer

Updated October 27, 2026. Hatchet and Avalanche are both MIT-licensed and both make multi-step work durable, so finished steps are not redone after a failure.[2, 4, 20, 25] Choose Hatchet if developers need a task queue and durable execution engine for their application, with fine-grained concurrency, rate limits and priorities, SDKs in Python, TypeScript, Go and Ruby, and a generally available cloud with a free tier.[1, 3, 6, 8] Choose Avalanche if the job is a long, multi-step, document-heavy business workflow: you set the few steps, checks and approvals that matter, agents do the work inside each step, and every run keeps typed step outputs and versioned files your team can compare across runs and versions, with instant rollback on Avalanche Cloud.[21, 23, 25]

Key takeaway

Choose Avalanche when agents should do the work inside a few business steps you control, often over PDFs, Word files and spreadsheets.

Choose Hatchet when developers need a fast, generally available task queue with fine-grained concurrency and rate limits for their application today.

  • Avalanche wins on level of abstraction: you set the steps, checks and approvals; an agent does the work inside each step instead of task code you write for every step.
  • Avalanche wins on versions and rollback: Avalanche Cloud rolls back to any previous version in one click, with no rebuild, and runs keep their version; in Hatchet, new runs use the latest workflow version registered, so rolling back means redeploying workers with the old code.
  • Hatchet wins on availability today: Hatchet Cloud is generally available with a free tier, and the open-source edition is fully featured; Avalanche Cloud is on a waitlist.

By , CTO, Trampoline AI. Updated · Facts checked · Published by the Avalanche team. Avalanche competes with Hatchet.

01 / The core differenceAvalanche vs Hatchet

A task queue for your application, or agentic workflows for your business.

Hatchet gives developers a durable task queue: tasks, DAGs and durable tasks with retries, concurrency control and a real-time dashboard.[1, 2] Avalanche works a level higher: you describe the outcome and the few steps, checks and approvals that matter, and agents do the work in between, with each step checkpointed.[21, 25]

Hatchet is infrastructure for developers: it queues, schedules, retries and observes the tasks your application code defines.[1, 8] Avalanche is built for the work agents do for a business: you define the few steps, checks and approvals that matter, agents do the work in between, and every run keeps its version, its typed step outputs and the files each step changed.[21, 23, 25] Both record every run; Hatchet's dashboard and traces help a developer debug tasks, while Avalanche's step outputs let a team compare the results of agent work across many runs and versions.[15, 25]

02 / At a glanceChecked 2026-10-27

Avalanche and Hatchet, side by side.

Avalanche and Hatchet compared, checked 2026-10-27
CriterionAvalancheHatchetEdge
Level of abstractionCoarse business steps: you set the steps, checks and approvals, and an agent does the work inside each step.[20, 25]A developer writes each task, DAG or durable task as code; agent loops are durable tasks that spawn child runs.[1, 29]Avalanche edge
Agents and skills inside stepsAgent steps can run Claude through the Claude Agent SDK and load Agent Skills (SKILL.md); the Avalanche skill drafts a full workflow from a described outcome.[22, 26, 27]A cookbook lets a Claude Agent SDK process call Hatchet tasks as tools in a trusted environment; hatchet skills install teaches coding agents the Hatchet CLI.[12, 13]Avalanche edge
Document lineageAvalanche Cloud: every run gets a versioned workspace; each step's file changes are committed as a new version, so you can see which step changed a file, compare versions and download any step's files; agent steps get PDF, Word and spreadsheet skills.[23, 25]Task inputs and outputs are payloads; the PDF cookbook passes the file as base64 input, and parsing is code you bring.[14, 23, 25]Avalanche edge
Inspecting and comparing runsAvalanche Cloud: typed inputs and outputs at every step checkpoint, in one run history filtered by status, flow and deployment; a web console with a live DAG of every flow and run, the inputs and outputs of any step in one click, and the full console in your terminal, so you can compare a step across many runs and versions.[21, 25]A real-time dashboard with run inputs, outputs, traces and searchable logs, plus a terminal UI; run history kept 1 day free, 7 days pay-as-you-go, 90+ days on Custom.[3, 15, 33]Different jobs
Versions and rollbackAvalanche Cloud: each deploy is a frozen version of the whole workflow, mapped to an exact git commit, with locked, hash-verified dependencies; running workflows finish on the version they started on; instant rollback to any previous version from the CLI or console, no rebuild; an audit trail of who promoted or rolled back.[25]No rollback action: new runs use the latest workflow version registered, and a version is added when a worker starts with a changed definition. Rolling back means redeploying workers with the old code (on Managed Compute, a new build from Git). A run keeps its task graph, but tasks go to whichever worker serves that task name.[19, 35, 37, 38, 39, 40, 41]Avalanche edge
Throughput and queue controlsAvalanche Cloud: FIFO queueing when over capacity, with organization and worker-pool limits; run creation p50 1.2 s, measured October 27, 2026 on v0.6.6.[25]Concurrency keys, static and dynamic rate limits, priorities, worker slots and affinity. Hatchet reports load tests to 10,000 tasks per second and dispatch often under 50 ms. The figures on each side time different things, and there is no head-to-head benchmark.[2, 6, 7, 8]Different jobs
Availability and self-hostingSelf-hostable MIT-licensed runtime with operator, web console, REST API and terminal UI, more limited than Cloud; Avalanche Cloud is on a waitlist.[20, 24, 28]Hatchet Cloud is generally available with a free tier in four regions; the MIT-licensed open-source edition is fully featured and self-hosts on PostgreSQL.[1, 3, 31]Hatchet edge
Schedules, webhooks, CLI and APIAvalanche Cloud: cron schedules and webhooks declared in your workflow code, following rollbacks and promotions; webhook URLs never change on redeploy, redeliveries are deduplicated for 7 days, and idempotency keys return the same run for 7 days; a typed REST API with an OpenAPI spec, and a CLI.[25]Crons in code, SDK or dashboard; one-time scheduled runs; events; webhooks with Stripe, GitHub and Slack presets; idempotency keys; a REST API, a CLI and a local MCP server.[9, 10, 13, 36]Similar
Identity and accessAvalanche Cloud: SSO; organization roles and per-action permissions, managed in your organization's settings; scoped, encrypted secrets; agents see only what each step passes them (its inputs, its files and the secrets it declares); steps run non-root in a sandbox, cut off from cloud metadata and private networks; an audit trail of promotions and rollbacks; no spend limits yet.[25]Tenant roles from Owner to read-only Viewer and per-member payload visibility; SSO on Custom plans; audit logs of member, token and run actions; SOC 2 Type II on every plan; HIPAA BAA on Custom.[3, 16, 17, 18]Hatchet edge
03 / Where each one winsOne claim per card

Where each one wins

Where Avalanche wins

  • Higher-level workflows

    You set the few business steps, checks and approvals that matter; an agent does real work inside each step. Python steps keep exact rules and arithmetic.[20, 25]

  • From a skill to a workflow

    Install the Avalanche skill with npx skills add Trampoline-AI/avalanche, describe the outcome in your coding agent, and it drafts the workflow with you.[22]

  • Every step versions your documents

    Avalanche Cloud gives every run a versioned workspace. Pass one document or hundreds; each step works on the files directly, and every step's changes are committed as a new version of the workspace, so you can see which step produced or changed each file, compare versions, and download any step's files. Document lineage, the way data teams have table lineage. Agent steps get PDF, Word and spreadsheet skills.[23, 25]

  • Versioned deploys and instant rollback

    On Avalanche Cloud each deploy is a frozen, checksum-verified version of the whole workflow, mapped to an exact git commit. Running workflows finish on the version they started on, and instant rollback makes any previous version live again in one click or one command, with no rebuild. Avalanche Cloud is on a waitlist.[25, 28]

  • Runs your business team can compare

    Each step is a business result with typed outputs, such as an RFP answer or a contract finding. Avalanche Cloud keeps every run by workflow and version, so you can compare one step across many runs.[21, 25]

  • Checkpointed Claude agent steps

    Agent steps can run Claude through the Claude Agent SDK and load Agent Skills (SKILL.md), each sandboxed and checkpointed like any other step.[26, 27]

Where Hatchet wins

  • Fine-grained queue controls

    Concurrency keys, static and dynamic rate limits, priorities, worker slots and affinity. Hatchet reports load tests to 10,000 tasks per second and dispatch often under 50 ms with PostgreSQL storage.[2, 8]

  • Fine-grained queue controls

    Concurrency keys with round-robin or cancel strategies, static and per-user rate limits, priorities, and worker slots keep load fair and within limits.[2, 6, 7]

  • Four language SDKs

    Tasks and workflows are code in Python, TypeScript, Go or Ruby, and workers run anywhere containers run.[1, 15]

  • Generally available, open source and self-hostable

    Hatchet Cloud has a free tier of 1M task runs and 1M events a month, and the MIT-licensed open-source edition is fully featured, built on PostgreSQL.[1, 3]

  • Developer observability

    A real-time web UI with alerting, logs with full-text search, a built-in OpenTelemetry collector, Prometheus metrics and a terminal UI.[2, 15, 33]

  • Access controls and audit logs

    Tenant roles from Owner to read-only Viewer, per-member payload visibility, SSO on Custom plans, audit logs, and SOC 2 Type II on every plan.[3, 16, 17, 18]

04 / Detailed comparison4 buyer questions

What changes for your team.

01

A task queue or business workflows?

Buyer question

Is the work part of your application's backend, or a business process your team repeats every week?

Hatchet is a platform for background tasks, AI agents and durable workflows in your application. You write tasks as functions, compose them into DAGs or durable tasks, and run workers that pick them up; Hatchet handles queueing, retries, scheduling and monitoring.[1, 2] It fits high-volume backends: ingestion, fan-out over thousands of items, or an agent loop inside your product.[29]

Avalanche is built for long, multi-step business work that agents do, such as drafting RFP answers from past proposals or reviewing a stack of contracts. You set the few steps, checks and approvals that matter, agents do the work in between, and simple workflows fit the same model. On Avalanche Cloud, workflows run for hours or days, and an owner approves results before they go out.[25]

02

Durability and recovery: close on both sides

Buyer question

What happens when a step fails halfway through a long run?

In Hatchet, a DAG persists each task's status and result, so a retry does not re-run the parts that succeeded. Durable tasks checkpoint every wait and child spawn and replay from their event log, which means the code between checkpoints must be deterministic. Tasks run at least once, so Hatchet's docs ask for idempotent task code; retries support exponential backoff.[4, 5, 8, 30]

Avalanche Cloud records completed steps durably and does not re-run them after a crash; failed steps retry, and uncertain outcomes are held instead of blindly retried. Steps are ordinary Python or agent steps, with no determinism rule to follow between checkpoints.[21, 25]

03

Documents, agents and skills

Buyer question

How much do you want to build around the agent yourself?

In Hatchet, documents travel as task payloads: the PDF pipeline cookbook sends the file as base64 input and parses it with a library you choose.[14] Agents are code you write: a durable task loops by spawning child runs, and a Claude Agent SDK cookbook lets Claude call Hatchet tasks as tools. Durable event waits are the building block for human-in-the-loop steps.[11, 12, 29]

In Avalanche, files move between steps, and on Avalanche Cloud each step's file changes are committed as a new version of the run's workspace. Agent steps declare typed inputs and outputs and get PDF, Word and spreadsheet skills.[21, 23] The Avalanche skill drafts the workflow from an outcome you describe in your coding agent.[22] Agent steps can also run Claude through the Claude Agent SDK and load Agent Skills (SKILL.md).[26, 27]

04

Running it in production

Buyer question

How mature does the managed service need to be today, and what do you need to see about each run?

Hatchet has the edge on availability and scale. Hatchet Cloud is generally available with a free tier, regions in the US, Europe and Australia, SOC 2 Type II on every plan, and SSO, audit logs and HIPAA on Custom plans; the open-source edition self-hosts on PostgreSQL with the same programming model.[3, 31] Its dashboard, OpenTelemetry traces and Prometheus metrics are built for developers running tasks at scale.[2, 15] Rollback is where the models differ. Hatchet runs the latest workflow version your workers registered, so undoing a bad release means redeploying workers with the previous code, or on Managed Compute a new build from Git.[37, 38, 40]

You can self-host the MIT-licensed Avalanche runtime, with its operator, web console, REST API and terminal UI, on your own infrastructure; it is more limited than Avalanche Cloud, which runs the production infrastructure for you.[20, 24] Avalanche Cloud adds SSO and organization roles, versioned deploys, instant rollback of the whole workflow to any previous version with no rebuild or worker redeploy, step checkpoints, managed queueing and one run history for every deployed workflow, and is open through a waitlist.[25, 28] Both record each run; the difference is what a step is. A Hatchet task is a function a developer wrote, so its trace helps debug code. An Avalanche step is a business result an agent produced, with typed outputs and versioned files, so a team can compare that step across hundreds of runs and see which outputs drift or which version changed them.[21, 25]

05 / FAQDirect answers

Common questions.

Is Avalanche an alternative to Hatchet?
Partly. Both are MIT-licensed and make multi-step work durable.[2, 20] Hatchet is a task queue and durable execution engine for application code at high volume; Avalanche runs business workflows where agents work inside a few steps you set, with versioned deploys and document lineage on Avalanche Cloud.[1, 25]
Does Hatchet support workflow rollback?
No. Hatchet stores a new workflow version whenever a worker registers a changed definition, and new runs always use the latest one; there is no console, CLI or API action to make an earlier version live again, and the version API only reads versions.[19, 35, 37, 38, 39] Going back means redeploying workers with the older code; on Hatchet Managed Compute, deploys are builds triggered by commits to your branch.[40] A run that has started keeps its task graph, but its tasks are sent to whichever worker serves that task name, and for breaking changes to durable tasks Hatchet recommends a new task definition and letting old work drain.[41] Avalanche Cloud rolls a workflow back to any previous version in one click or one command, with no rebuild, and running workflows keep the version they started on.[25]
Can Hatchet run AI agents?
Yes. Hatchet's docs describe agents as durable tasks that spawn child runs for each step or tool call, and cookbooks show the Claude Agent SDK and OpenAI Agents SDK calling Hatchet tasks as tools.[12, 29]
Is Hatchet open source?
Yes. Hatchet is MIT-licensed and its open-source edition is fully featured; it self-hosts on PostgreSQL.[1, 2] The Avalanche runtime is also MIT-licensed on GitHub.[20]
How is Hatchet billed?
Pay-as-you-go: the first 1M task runs and 1M events each month are free, then $30 per million task runs and $5 per million events; retries are not billed, and storage, log volume and users above the included amounts are extra. Custom plans add volume discounts, SLAs, SSO, RBAC and audit logs. Prices as of October 27, 2026.[3]
Can I see which step changed a document?
Yes, on Avalanche Cloud. Every run gets a versioned workspace, and each step's changes are committed as a new version, so you can see which step produced or changed each file, compare versions and download any step's files.[25] Hatchet passes task inputs and outputs as payloads; its docs do not define a file workspace carried between steps.[14]
06 / SourcesChecked 2026-10-27

Public primary sources.

Every source was fetched on the date shown. Prices and plans change; check the vendor before you buy.

  1. [1]What is Hatchet? (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/home
  2. [2]hatchet-dev/hatchet repository and README (MIT License)Hatchet on GitHub · Checked https://github.com/hatchet-dev/hatchet
  3. [3]Hatchet pricingHatchet · Checked https://hatchet.run/pricing
  4. [4]DAGs as durable workflows (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/directed-acyclic-graphs
  5. [5]Durable tasks (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/durable-tasks
  6. [6]Concurrency control (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/home/concurrency
  7. [7]Rate limits (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/rate-limits
  8. [8]Architecture and guarantees (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/architecture-and-guarantees
  9. [9]Recurring runs with cron (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/cron-runs
  10. [10]Webhooks (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/webhooks
  11. [11]Durable event waits (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/durable-event-waits
  12. [12]Claude Agent SDK cookbook (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/cookbooks/hatchet-claude-agent-sdk-trusted-env
  13. [13]Using coding agents (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/using-coding-agents
  14. [14]PDF pipeline cookbook (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/cookbooks/pdf-pipeline
  15. [15]Hatchet home pageHatchet · Checked https://hatchet.run/
  16. [16]User roles (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/user-roles
  17. [17]Single sign-on (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/single-sign-on
  18. [18]Audit logs (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/security/audit-logs
  19. [19]Workers (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/workers
  20. [20]Avalanche repository and README (MIT License)Trampoline AI · Checked https://github.com/Trampoline-AI/avalanche
  21. [21]Results, cancellation, and reruns (Avalanche docs)Avalanche docs · Checked https://avalanche-docs.docsalot.dev/build-workflows/results-and-reruns
  22. [22]Build a workflow with the Avalanche skill (Avalanche docs)Avalanche docs · Checked https://avalanche-docs.docsalot.dev/build-workflows/build-with-skill
  23. [23]Skills for agent steps: PDF, DOCX and spreadsheet (Avalanche docs)Avalanche docs · Checked https://avalanche-docs.docsalot.dev/agent-steps/skills
  24. [24]Architecture and current limits (Avalanche docs)Avalanche docs · Checked https://avalanche-docs.docsalot.dev/reference/architecture-and-limits
  25. [25]Avalanche and Avalanche Cloud overviewAvalanche · Checked https://avalanche.run/
  26. [26]Pull request #124: Claude Agent SDK agent stepsTrampoline AI on GitHub · Checked https://github.com/Trampoline-AI/avalanche/pull/124
  27. [27]Pull request #136: Agent Skills (SKILL.md) in agent stepsTrampoline AI on GitHub · Checked https://github.com/Trampoline-AI/avalanche/pull/136
  28. [28]Avalanche Cloud waitlistAvalanche · Checked https://cloud.avalanche.run/waitlist
  29. [29]Child spawning (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/child-spawning
  30. [30]Retry policies (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/retry-policies
  31. [31]Cloud vs OSS (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/cloud-vs-oss
  32. [33]Using the Hatchet TUI (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/reference/cli/tui
  33. [35]Workflows client (Hatchet Python SDK reference)Hatchet · Checked https://docs.hatchet.run/reference/python/feature-clients/workflows
  34. [36]Idempotency (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/idempotency
  35. [37]New runs select the latest workflow version (triggers.sql, ListWorkflowsByNames)Hatchet on GitHub · Checked https://github.com/hatchet-dev/hatchet/blob/0785cd67cee9b77d6e4622aa52cafcba00c0a4fc/pkg/repository/sqlcv1/triggers.sql#L43-L59
  36. [38]A new workflow version is stored when a registered definition changes (workflow.go, createWorkflowVersionTxs)Hatchet on GitHub · Checked https://github.com/hatchet-dev/hatchet/blob/0785cd67cee9b77d6e4622aa52cafcba00c0a4fc/pkg/repository/workflow.go#L943-L954
  37. [39]Workflow version API is read-only (GET, latest by default)Hatchet on GitHub · Checked https://github.com/hatchet-dev/hatchet/blob/0785cd67cee9b77d6e4622aa52cafcba00c0a4fc/api-contracts/openapi/paths/workflow/workflow.yaml#L244-L268
  38. [40]GitOps with Hatchet Managed Compute (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/compute/git-ops
  39. [41]Migrating from Temporal to Hatchet: versioning and determinism (Hatchet docs)Hatchet · Checked https://docs.hatchet.run/v1/from-temporal-to-hatchet
Related comparisonsView all comparisons
See the runtime for yourselfOpen source · MIT

Our code speaks for itself.

The Avalanche runtime is open source under the MIT License. Read the code, run the starter project, and see how it fits the work your team repeats.