Manyfold
Github
Manyfold · agent hosting & delivery

Host your agents.
Your work, multiplied.

Claude Code, Codex and your own agents, always on and delivering into your product.

Scroll · see how it works ↓
01 · Hosting

Agents live in runtimes.

Not tied to your laptop. Every agent gets a persistent home.

  • Any framework · Claude Code, Codex, Gemini CLI, and more
  • Any machine · managed sandboxes, cloud computers, your own hardware
  • Already built? · plug in Dify, Langflow and external agents
Run yours on a cloud machine
02 · One workspace

One place runs all of them.

Whatever the framework, every agent shows up the same way.

  • Sessions, files, terminal · real workspaces, backed up
  • Models per agent · managed or bring your own provider
  • Usage per turn · tokens, cost and latency, always visible
03 · Wherever you work

Same agent. Every surface.

One capability, reachable from everywhere you already work.

  • Your product · call it over an OpenAI-compatible API
  • Team chat & terminal · Lark, Slack, WhatsApp, or the mf CLI
  • Your schedule · automations run it, results get delivered
See every app it reaches
04 · The point

Your work, multiplied.

PRs reviewed while you sleep. Reports drafted before standup. Tickets triaged before you open the queue. Your product and your team, running on more hands than you have.

Works with

Bring what you already use.

Frameworks keep their native behaviour. Channels reach your team where they are. Runtimes are yours to pick.

Frameworks
Claude CodeCodexGemini CLIPiAntigravity CLIOpenclawHermesNarraNexusDifyLangflowA2A
Channels
LarkSlackDiscordTelegramWhatsAppMatrixWeChatLINEiMessageLinearGitHub
Runtimes
Stateful sandboxCloud computerYour own machineExternal servicesStateful sandboxCloud computerYour own machineExternal servicesStateful sandboxCloud computerYour own machineExternal services
Observability

Every run,
on the record.

Transcripts, cost and permissions for every turn — including the runs that happened while you were asleep.

cc-review#42h
readsrc/auth/session.ts
edit3 files
test18 passed
PR #412
Transcript.

Every turn kept — the prompt, the tool calls, the files it touched. Pick the session back up weeks later.

claude-sonnet-5$0.120
gpt-5-codex$0.052
gemini-2.5-pro$0.011
month to date$12.40
Cost.

Tokens, latency, model and price per turn, across every provider you use.

editsrc/auth/session.tsAllowed
runpnpm testAllowed
runrm -rf node_modulesDenied
Control.

Every permission an agent asked for, and what you answered, on the same record.

Manyfold Solutions

Need something built? Live in days.

Websites, internal systems, new features and AI assistants, built by our engineers at a price fixed before we start.

See AI solutions
Plans

Start free. Scale when ready.

Every plan runs the same platform. Pick the limits that fit how many agents you keep online.

Free

Great for getting started and seeing what Manyfold can do.

Free
  • 3 Sandbox agents
  • 2 Always-online agents
  • 5 sandbox active hours / month
  • 1 always-online runtime
  • 2 message channels
  • 5,000 API calls / month
Hobby

For solo builders running a few agents in production.

$9/ mo
  • 10 Sandbox agents
  • 6 Always-online agents
  • 20 sandbox active hours / month
  • 3 always-online runtimes
  • 8 message channels
  • Unlimited API calls
PlusPopular

For small teams collaborating across many agents.

$19/ mo
  • 25 Sandbox agents
  • 18 Always-online agents
  • 60 sandbox active hours / month
  • 10 always-online runtimes
  • 25 message channels
  • 180 days message history
Pro

For teams running production workloads at scale.

$49/ mo
  • 75 Sandbox agents
  • 54 Always-online agents
  • 200 sandbox active hours / month
  • 30 always-online runtimes
  • 100 message channels
  • Unlimited message history & automation runs

Prices in USD. Cancel anytime. Need a custom tier? Reach out and we will size it with you.

Questions

The ones that decide it.

If you are weighing this against building it yourself, start here.

No. Each framework runs in its native form, with its own execution model, model configuration and workspace semantics. Manyfold is the layer around them: runtime, workspace, orchestration, delivery and metering. When a framework ships something new, you get it the same day.
Yes. Open "Use in your agent" at the top of this page and paste the prompt into your agent. It installs the mf CLI, opens your browser once so you can sign in or create an account, and adds the Manyfold plugin. Works with Claude Code, Codex and any agent that can run shell commands.
A VM gives you compute. Manyfold gives you the part you would build next: framework bootstrap, persistent sessions, files and terminal, model configuration, skills and MCP, channels, automations and per-turn usage, for every agent, in one place.
No. Connect your own laptop or server with the mf daemon, or plug in Dify, Langflow and other supported services as adapted work units. External runtimes keep their own capability boundaries, and we say where those boundaries are rather than pretending they do not exist.
Your hosted agent behind an OpenAI-compatible endpoint, so existing SDKs work unchanged. Calls join the same session, permission and usage system as the web workspace. It is an agent you manage, not a bare model proxy.
Most developers start in the web workspace, then wire the API into a product or a channel into team chat. Ops-minded teams often start with a schedule and a channel. Every surface reaches the same agents, so the order costs you nothing.