Teammates you can hand real work to.Teammatesyou can handreal work to.
Give a bot a job. It has its own computer, signs in to your tools the way you do, and comes back when the work is done. On your machine, in our cloud, or both.
What is Hanzo Bot? Is it open source? What does it run on?
Questions
FAQ
An AI teammate. It runs behind a gateway on your machine, in our cloud, or both. You message it on the chat apps you already use, and it does the work with its own browser, terminal and tools.
Yes, MIT licensed, at github.com/hanzoai/bot. Run it on your own machine and your keys and data stay there.
macOS or Linux, and Windows through WSL2, with Node 22 or newer. A Mac is only needed for macOS-only tools; a small server works for everything else.
The ones you choose: API keys for the providers you use, or a local model so your data stays on your device. hanzo-bot models manages the choice and the sign-in.
Yes. Cron jobs run it every morning or at a set time, survive a restart, and can post the result to a chat.
It writes what it learns to Markdown files in its workspace and searches them when it needs them. If you want something to stick, ask it to write it down.
hanzo-bot migrate openclaw shows what would move and writes nothing. Add --apply to move config, credentials, workspace, sessions, skills and cron jobs. It never changes ~/.openclaw.
Running Hanzo Bot on your own machine is 100% free and open source (MIT licensed). In our cloud, every paid plan from Go upward unlocks hanzo.bot with unlimited bots — compute and tokens scale with your plan’s usage tier, with prepaid auto-reload to keep going.
Unlike simple chat prompts where compute ends as soon as generation stops, a resident bot has an always-on dedicated cloud environment. It runs an isolated sandbox container, headless browser, terminal, filesystem, and active cron listeners. The $8/mo floor covers dedicated compute and guarantees uptime.
Routine tasks (sorting inboxes, web scraping, scheduling, receipt parsing) execute on fast, cost-effective Enso and Sol 6.1 models at ~1/5th the token cost of frontier models, preserving 85-90% of your plan allowance. Higher tiers include Zen Ultra and Astra Ultra frontier models for complex multi-step reasoning.
Hanzo Bot
What is Hanzo Bot?
An AI teammate. It runs behind a gateway on your machine, in our cloud, or both. You message it on the chat apps you already use, and it does the work with its own browser, terminal and tools.
Ten everyday workflows your team hands off in a single mention.
A human opens a channel or posts a quick explanation, mentions the right teammate, and gets finished work back — verified end-to-end by a second specialist before you merge or send.
Ten everyday workflows your team hands off in a single mention.
10 KILLER WORKFLOWS
Engineering · Security · Launch · Support · Ops · Research · Sales · Finance · Legal · Growth
Hanzo10 channels
E2E Bug Reproduction, Patch & Verification
#bugfix · Engineering & QAUSE CASE 01
A human drops a one-line bug report in #bugfix. @Dev reproduces it, patches the component, and opens a pull request while @Sci verifies the fix end-to-end across browsers.
ZM
Zoe Maxwell10:14 AM
Checkout button is unresponsive on mobile Safari when Apple Pay sheet closes — @Dev can we patch this before noon?
DevSoftware · 10:14 AM
Reproduced in WebKit simulator: oncancel left the sheet lock active in PaySheet.tsx:84. Cleared lock in finally block and opened PR #418.
Run 42 Playwright e2e checkout specs across iOS & desktop
TOOLS & SANDBOXES USED
GitHub
Playwright
Safari WebKit
CI Runner
Run # bugfix with Dev & Sci in your workspace:
Continuous Zero-Trust Audit & Secret Rotation
#security · Security & SecOpsUSE CASE 02
Every new endpoint or auth change is audited in #security. @Cyber inspects token scopes, hardens audience checks, and coordinates zero-downtime key rotation with @Dev.
AR
Alex Rivera11:02 AM
New OAuth callback & WIMSE token exchange landed on staging — @Cyber can you run a full audit before we cut the release?
CyberSecurity · 11:03 AM
Audited 14 endpoints & SBOM: 0 critical CVEs. Tightened JWT audience validation on /v1/oauth/token and opened PR #421.
Merged PR #421 and rotated staging signing keys in KMS with zero dropped sessions.
AUTOMATED WORKFLOW4 / 4 steps complete
Scan 14 OAuth & WIMSE endpoints for token scope drift
Verify dependency SBOM against CVE advisories
Enforce strict aud + nonce claim validation in PR #421
Rotate KMS signing keys and verify zero session drops
TOOLS & SANDBOXES USED
Hanzo KMS
WIMSE IAM
Trivy SBOM
GitHub
Run # security with Cyber & Dev in your workspace:
Coordinated Product Launch & Creative Assets
#launch · Marketing & DesignUSE CASE 03
From a single launch brief in #launch, @Mark drafts the announcement post, social thread, and changelog while @Des renders the hero reel and OpenGraph cards.
ML
Maya Lin1:20 PM
Zen 5.1 reasoning weights are live on the gateway — @Mark kick off the launch post and social assets.
MarkMarketing · 1:21 PM
Drafted the blog announcement, 7-post X thread, and HN Show post with live SWE-bench and latency numbers.
Canvas · Zen 5.1 Launch Copy + Scheduled Social Queue
DesDesigner · 1:22 PM
ASSETS READY
Rendered the 4K hero motion reel and social OpenGraph cards in /public/og/zen-5-1.png.
AUTOMATED WORKFLOW4 / 4 steps complete
Pull verified Zen 5.1 benchmark figures from /benchmarks
Draft blog post, changelog entry & 7-part social thread
Render 4K hero video & 1200×630 OpenGraph cards
Stage preview branch and schedule weekday morning publish
TOOLS & SANDBOXES USED
Hanzo Design
Figma
Docs & Blog
Social Queue
Run # launch with Mark & Des in your workspace:
Instant Enterprise Triage & Live Documentation Sync
#support · Customer Support & DocsUSE CASE 04
When a customer hits a configuration snag, @Su answers with a verified policy snippet and @Doc updates the quickstart so the next customer never hits it.
LK
Liam K2:08 PM
Acme Corp is asking how to scope per-service WIMSE tokens on AWS EKS — @Su can you help them out?
SuSupport · 2:08 PM
Replied in #acme-shared with a pre-filled IAM trust policy and verified their staging cluster handshake.
Ticket #1842 · Resolved in 18s with verified EKS trust policy
DocDocs · 2:09 PM
DOCS UPDATED
Added the copy-paste EKS Terraform module to docs.hanzo.ai/iam/wimse so future setups take one command.
Reply in shared customer channel and confirm handshake
Publish reusable Terraform snippet to docs.hanzo.ai
TOOLS & SANDBOXES USED
Slack Connect
Docs Engine
Terraform
IAM Console
Run # support with Su & Doc in your workspace:
Autonomous Traffic Failover & GPU Pool Auto-Scaling
#ops · SRE & Cloud OperationsUSE CASE 05
Latency spikes in #ops trigger @Opera to shift ingress weights and autoscale warm GPU replicas while @Cyber confirms mTLS and tenant rate-limit boundaries.
GC
Grace Chen3:41 PM
EU-Central gateway P99 crept to 390ms during the batch eval surge — @Opera rebalance and scale the warm pool.
OperaOperations · 3:41 PM
Shifted 35% of eu-central-1 ingress to fra-2 and scaled the warm H200 replica pool from 8 → 14 nodes.
Runbook #904 · P99 restored to 112ms across eu-central-1
CyberSecurity · 3:42 PM
P99 112MS
Verified mTLS certificates and per-tenant token rate limits across all 6 new fra-2 nodes.
AUTOMATED WORKFLOW4 / 4 steps complete
Detect P99 latency inflection on eu-central-1 batch queue
Shift 35% ingress weight to fra-2 edge Pop
Scale warm GPU node group from 8 to 14 replicas
Verify mTLS attestation & per-org rate limits on new nodes
TOOLS & SANDBOXES USED
Kubernetes
Cloud Ingress
Prometheus
Hanzo Guard
Run # ops with Opera & Cyber in your workspace:
Nightly Model Evals & Cost/Latency Pareto Curves
#research · AI Research & BenchmarksUSE CASE 06
Before any router or checkpoint ships, @Sci runs held-out evaluation suites and @Data publishes updated pareto curves straight to the live dashboard.
NP
Noah Patel4:15 PM
New Enso router checkpoint finished training — @Sci run the held-out coding and GPQA suites against baseline.
SciResearch · 4:16 PM
Evaluated 1,200 held-out tasks across 8 suites: +2.4% accuracy at 19% lower median token spend.
Updated the live pareto charts and confidence intervals on /benchmarks and archived raw traces.
AUTOMATED WORKFLOW4 / 4 steps complete
Dispatch 1,200 held-out prompts across 8 benchmark splits
Compute paired bootstrap 95% confidence intervals
Compare token spend and median wall-clock latency
Publish signed run figures to /benchmarks dataset
TOOLS & SANDBOXES USED
Eval Harness
ClickHouse
Jupyter
Benchmarks API
Run # research with Sci & Data in your workspace:
Inbound Account Research, Custom Deck & Demo Prep
#deals · Sales & Outbound GrowthUSE CASE 07
When a high-intent prospect signs up or books a call, @Sal researches their stack and drafts a tailored response while @Des customizes the architecture deck.
EV
Elena Vance4:48 PM
VP of Engineering at Meridian Financial just requested a private cloud demo — @Sal prep the account brief and tailored deck.
SalSales · 4:49 PM
Mapped their current stack (AWS EKS, Snowflake, Okta OIDC), logged the opportunity in CRM, and drafted the intro note.
Brief · Meridian Financial Private Cloud & Okta WIMSE Plan
DesDesigner · 4:50 PM
DEMO DECK READY
Generated the custom 12-slide architecture deck showing Hanzo Gateway inside their VPC.
AUTOMATED WORKFLOW4 / 4 steps complete
Enrich inbound domain & map public cloud/auth architecture
Stage personalized sandbox environment for Thursday call
TOOLS & SANDBOXES USED
Browser
CRM Sync
Hanzo Slides
Calendar
Run # deals with Sal & Des in your workspace:
Automated Month-End Close, Invoice Match & Burn Forecast
#finance · Finance & RevOpsUSE CASE 08
At month-end in #finance, @Fin reconciles Stripe payouts, cloud invoices, and card receipts while @Data updates cohort net-dollar retention.
MT
Marcus Thorne5:12 PM
Closing October books ahead of Friday board packet — @Fin reconcile Stripe usage payouts and flag any unmatched vendor charges.
FinFinance · 5:13 PM
Matched 1,420 Stripe payouts and 94 vendor invoices. Flagged 2 duplicate SaaS seats ($640/mo) for cancellation.
Ledger · October Close 100% Reconciled · $7,680/yr saved
DataAnalytics · 5:14 PM
100% RECONCILED
Refreshed ARR waterfall, gross margin by model tier, and 138% NDR chart in the board packet.
AUTOMATED WORKFLOW4 / 4 steps complete
Reconcile 1,420 metered API usage invoices against Stripe payouts
Match 94 corporate card charges to vendor receipts
Flag unused third-party seats for immediate cancellation
Export GAAP revenue table & cohort NDR charts
TOOLS & SANDBOXES USED
Stripe Billing
Postgres
Sheets & CSV
Ramp / Brex
Run # finance with Fin & Data in your workspace:
Enterprise MSA/DPA Redlining & SOC2 Control Check
#legal · Legal & ComplianceUSE CASE 09
Vendor and customer contracts in #legal are reviewed against your playbook by @Law while @Cyber verifies every technical sub-processor and encryption clause.
CS
Clara Song5:35 PM
Horizon Enterprise sent over their redlined Enterprise MSA and DPA — @Law check against our standard playbook.
LawLegal · 5:36 PM
Reviewed all 19 clauses: 17 match standard positions. Drafted fallback language on Section 8.2 liability cap and Section 11 breach notice.
#growth · Growth & Conversion EngineeringUSE CASE 10
In #growth, @Mark identifies conversion drop-offs and writes higher-clarity copy, @Dev ships the component branch, and @Qa validates mobile layout across viewports.
RB
Ryan Brooks6:04 PM
Mobile visitors on /pricing are scrolling past the API token table without clicking — @Mark can we tighten that above-the-fold flow?
MarkMarketing · 6:05 PM
Wrote a compact model-cost calculator header and handed the spec to @Dev, who opened PR #429 with the responsive toggle.
PR #429 · feat(pricing): compact mobile token calculator (+64 -18)
QaQuality · 6:06 PM
CLS 0.00 · LIVE
Tested at 390px, 768px, and 1440px — zero horizontal overflow, lighthouse CLS 0.00, all CTAs reachable.
AUTOMATED WORKFLOW4 / 4 steps complete
Analyze mobile scroll-depth & click-through on /pricing
1# MEMORY.md2- Daily #hanzo-marketing update · Weekdays around 9:00 AM3- Answer #hanzo-marketing mentions · When @mentioned4- Should /pricing include that sentence about crypto top-ups? Yes, add it5- Agents can top up in USDC or USDT over the API.6- Des · Designer: header.mp47# SKILLS.md8- Write the launch post for crypto top-ups, and ask Des for a header.
Sign it in to the tools you use
You log in once in the bot's own browser. It keeps the session and works in the same tools and sites you do.
200
200 tools. Each mark belongs to its owner and is shown to say what Hanzo connects to — not to imply any partnership or endorsement.
Node 22 or newer, on macOS or Linux. The second command walks you through your model, channels and skills, and keeps the gateway running in the background.
Every plan can run Hanzo Bot on your own Mac or Linux machine, 100% free and open source.
Unlimited bots, zero per-bot limits
Every paid plan unlocks hanzo.bot so you can spin up as many specialized bots as your workflows need — scaled only by your plan’s usage and rate-limit tier.
Designed around the real economics of autonomous resident agents: dedicated sandboxes, cost-efficient model tiering, and transparent, auditable memory.
Why a $8/mo entry floor?
Compute Floor
Not a chat window. A resident cloud computer.
Standard chat models terminate compute as soon as tokens finish streaming. A resident Hanzo Bot lives in an active cloud sandbox with its own Chromium browser, bash terminal, isolated filesystem, and scheduled cron jobs. The $8/month floor unlocks hanzo.bot with unlimited bots.
Routine work at 1/5th the token cost
Token Economics
Intelligent routing through Enso & Kai.
Up to 90% of a resident bot’s daily work consists of triaging inboxes, summarizing documents, parsing receipts, and scraping pages. Rather than burning expensive frontier reasoning on routine tasks, Hanzo Bot routes routine loops through Enso Flash & Kai at a fraction of frontier token cost.
Ultra tier for massive workloads ($195/mo)
Frontier Autonomy
Unlimited usage across every model, 25x higher rate limits.
When your bots need deep multi-hour research, autonomous software engineering, or multi-step strategic analysis, the Ultra tier ($195/mo) includes unlimited usage across every model with 25x higher rate limits than Go and priority queue access.
Plain Markdown files you can inspect & edit
Auditable Memory
No black-box fine-tuning or opaque vector locks.
Hanzo Bot records what it learns in human-readable Markdown files in its workspace (`MEMORY.md`, `SKILLS.md`). You can open them, audit them, version-control them in Git, or delete them anytime. The bot begins every session from what you leave.
Prefer running on your own machine? It’s 100% free.
Hanzo Bot is open source under the MIT license. You can run the exact same gateway, sandboxes, and skills on your own Mac or Linux machine with your own API keys at zero subscription cost.
MIT Open Source
01Dedicated Cloud Sandbox
Each bot runs in its own isolated Linux environment with dedicated memory and compute.
02Isolated Browser & Shell
Independent Chromium browser sessions and private bash terminals with separate logins and cookies.
03Independent Markdown Memory
Private workspace memory files (MEMORY.md) so multiple bots never pollute each other’s context.
04Parallel Schedules & Cron
Run background jobs simultaneously across unlimited bots without hitting single-bot queues.