Agenvoy
Self-hosted AI agent harness that auto-builds, sandbox-tests, and shares its own tools over MCP
Pick Agenvoy if you want an agent that gets more capable over time and shares that capability with your other coding agents over MCP. The auto-build-then-sandbox-test loop plus server mode is hard to replicate with a hosted assistant, and bubblewrap/sandbox-exec isolation with per-tool file snapshots is what makes it safe to leave running. The friction is real: CLI-first install via a curl script, Linux and macOS only, no managed hosting, and you supply your own provider keys. Compare against hosted agent platforms that run the loop for you but keep your logins on their VM.
Verified 2d ago · liveness 67/100 · cite: rightaichoice.com/tools/agenvoy
- Developers who want an always-on self-hosted agent that builds and reuses its own tools instead of resetting each
- Teams sharing one sandboxed tool library across Claude Code, Codex, OpenCode or Cursor via MCP server mode
- Privacy-conscious users running fully local through Ollama, LM Studio or vLLM so prompts never leave the machine
- Power users automating recurring jobs with cron, skill binding and Telegram or Discord push
- Non-technical users wanting a GUI-first, no-code, zero-install experience
- Teams that need managed hosting, vendor SLAs or enterprise support contracts
- Anyone needing a polished mobile app or dedicated desktop client
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Agenvoy if you want a zero-install, GUI-first assistant someone else hosts and supports — it is a daemon you install, tune and keep running yourself.
Keeping the daemon healthy is on you: with no vendor-managed hosting, a crashed or stopped process means your schedules and Telegram pushes simply stop firing.
Agenvoy is AGPL-3.0 open source, with commercial licensing available for closed-source use, so cost is mostly your own provider API spend plus the machine and time to run the daemon. That puts it well below hosted agent platforms that bundle infrastructure and support into a subscription, and in the same bracket as other self-hosted harnesses where you supply the model endpoints yourself.
In short
Agenvoy — Self-hosted AI agent harness that auto-builds, sandbox-tests, and shares its own tools over MCP. Best for Developers who want an always-on self-hosted agent that builds and reuses its own tools instead of resetting each, Teams sharing one sandboxed tool library across Claude Code, Codex, OpenCode or Cursor via MCP server mode, Privacy-conscious users running fully local through Ollama, LM Studio or vLLM so prompts never leave the machine. Free to use.
What's new in Agenvoy
Checked yesterdayAcross the latest 2 updates: 2 changelog entries.
Agenvoy v1.0.0 — first formal release, AGPL-3.0 with commercial licensing
The first formal release moved the project to AGPL-3.0 with commercial licensing available and documented the twelve provider entries, MCP client and server modes, and the memory and sandbox architecture.
v0.34.3 — TUI /voice toggle removed, voice moved to the browser
The TUI's /voice toggle was removed; hands-free voice now runs in the web dashboard via a VAD mic loop with wake words and interruptible spoken replies.
What people actually say about Agenvoy — is it worth it?
We scanned public community sources for Agenvoy on Sep 14, 2026 and could not establish that the discussion we found is about this tool rather than something else sharing its name. Only 4 of the posts we fetched could be positively tied to Agenvoy. Rather than publish a sentiment score built on the wrong subject, we publish nothing here and re-run the scan.
Viability Score
How well maintained and how widely used is Agenvoy? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Auto tool generation — finds an API, writes the script and schema, runs test_tool, registers it for reuse
- MCP server mode — exposes built-in, script, API and extension tools to Claude Code, Codex, OpenCode, Cursor over stdio
- MCP client mode — connects external MCP services over stdio or HTTP/SSE and auto-registers their tools
- Realtime voice in the browser dashboard via Gemini Live — hands work to the chat agent and narrates progress
- Interruptible voice conversation with wake-word detection — cut in mid-sentence; every request lands in the chat transcript
- Dispatcher routing across eleven provider entries with best-model selection per task
- Any OpenAI-compatible /v1 endpoint — Ollama, LM Studio, vLLM or a self-hosted gateway
- Cron and one-shot scheduling with skill binding, fsnotify hot-reload, auto-disable on failure
- Three-tier memory — context window, semantic vector search, full-text SQLite archive
- Cross-session error memory carrying failed-tool fixes forward across sessions
- Self-improvement — on tool failure the agent rewrites it, sandbox-tests it, retries and persists the fix
- OS-native sandboxing — bubblewrap on Linux, sandbox-exec on macOS, network off unless approved per call
- File snapshotting and rollback for a single version or an entire task
- 31 built-in tools across file, web, execution, coordination, state/memory and support groups
- Loadable markdown skill packs with YAML frontmatter, triggered by slash command, natural language or cron
About Agenvoy
Agenvoy is a Taiwan-developed, open-source (AGPL-3.0, commercial licensing available) AI agent harness from Pardn Chiu that runs as a single Go binary daemon on the computer you already own. Ask for something in plain language; when no tool covers the request, the agent detects the gap, searches public APIs, writes the script and schema, verifies it with test_tool inside an OS-native sandbox (bubblewrap on Linux, sandbox-exec on macOS, network off unless you approve it), then registers the tool for reuse. Next time it is already there. Sessions, memory and schedules live in SQLite under ~/.config/agenvoy and API keys sit in the OS keychain, used only by the local daemon. A dispatcher routes tasks across eleven provider entries behind one interface, plus any OpenAI-compatible /v1 endpoint such as Ollama, LM Studio or vLLM. MCP server mode gives Claude Code, Codex, OpenCode and Cursor your whole sandboxed tool library — and those agents can auto-build new tools that everyone instantly shares; client mode pulls in external MCP services over stdio or HTTP/SSE. A three-tier memory stack, cross-session error memory, cron scheduling with skill binding and per-tool file snapshotting round it out. Reach the same agent from a TUI, the embedded browser dashboard at http://127.0.0.1:17989, a Telegram bot, a Discord bot or an OpenAI-compatible REST endpoint. It is for developers and power users who want an always-on assistant that accumulates capability without handing files, logins or workflows to a hosted service.
Behind the Verdict
Agenvoy's pitch is that a chat gives you an answer while work needs a result, and the architecture backs that up. The centrepiece is auto tool generation: when no registered tool matches a request, the agent detects the gap, searches public APIs, fetches docs, generates a script and schema, verifies it with test_tool in the sandbox, then persists it. The scraped homepage walks a weather example end to end and a TSMC stock report scheduled with cron and pushed to Telegram. The parts that make this trustworthy rather than merely clever are the guardrails: commands run under bwrap on Linux or sandbox-exec on macOS with network off unless approved for that call, management endpoints answer localhost only, credentials stay in the OS keychain and the browser page never receives them, and every file a tool touches is snapshotted so you can roll back a single version or an entire task. Cross-session error memory carries failed-tool fixes into later sessions, so a broken tool tends to stay fixed. The second pillar is MCP. Server mode exposes 31 built-in tools plus script, API and extension tools over stdio to Claude Code, Codex, OpenCode and Cursor — one mcpServers entry in claude_desktop_config.json — and client mode connects external MCP services over stdio or HTTP/SSE with auto-registration. KuraDB registers as an MCP server for keyword-plus-semantic RAG over personal documents. Modelling is deliberately vendor-neutral: a dispatcher routes across eleven provider entries within one interface and will also talk to any OpenAI-compatible /v1 endpoint, so a fully local setup through Ollama, LM Studio or vLLM is viable. The docs enumerate surfaces beyond the dashboard — REST API, model and session endpoints, channel endpoints for Telegram and Discord, MCP endpoints, a TUI, and realtime voice in the browser via Gemini Live with hands-free wake words and interruptible replies. The honest weaknesses: this is a daemon you install, tune and keep running, with no vendor-managed SLA and no hosted copy behind it; install is a curl script, documented CLI-first; sandboxing exists only for Linux and macOS; providers must be wired through the dispatcher before anything works; and the extension surface (skill packs with YAML frontmatter, scripts, API tools, subagents, external MCP servers) is where most of the value lives, so it rewards tinkering and punishes anyone looking for a no-code GUI. Compare against hosted agent platforms that handle infrastructure but run your agent on the vendor's VM with your logins inside it.
Researching Agenvoy? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Agenvoy actually fits — and what changes day-one when you adopt it.
Adds one mcpServers entry to claude_desktop_config.json pointing at the local Agenvoy daemon, then asks for live data no existing tool covers.
Outcome: Agenvoy builds the script, runs test_tool in the sandbox and registers it, and every connected MCP agent can immediately call the same sandboxed tool.
Points the dispatcher at a local Ollama endpoint, binds a skill to a cron schedule, and routes results to a Telegram bot.
Outcome: Recurring work runs under bwrap with network off unless approved, prompts never leave the machine, and results are pushed outward without opening inbound ports.
Registers KuraDB as an MCP server while chatting in the TUI, then hits a tool failure mid-session.
Outcome: Keyword and semantic RAG search over local documents works from the same interface, and the agent rewrites, sandbox-tests and retries the failed tool, carrying the fix into later sessions.
Use Cases
- Automate a daily TSMC stock report at 08:00 via cron bound to a skill, with results pushed to Telegram.
- Ask for the weather in Taipei and have the agent build, test and register a fetch_weather script on the fly.
- Share your entire sandboxed tool library with Claude Code or Codex by adding one mcpServers entry to claude_desktop_config.json.
- Run RAG over local documents through KuraDB registered as an MCP server while chatting in the TUI or Telegram.
- Write a code-review skill pack that invokes subagents for parallel review, at most three legs at a time.
- Connect an external GitHub MCP server in client mode and let its tools auto-register for the agent immediately.
- Talk to the dashboard hands-free with a wake word and interruptible spoken replies while the full transcript stays in chat.
- Roll back a bad tool run using file snapshotting instead of reverting the whole machine.
Models Under the Hood
as of 2026-10-15
Limitations
- Self-hosted: Agenvoy runs as a daemon on your own machine (installed via a one-line curl script) and keeps sessions, memory and schedules in SQLite under ~/.config/agenvoy, with credentials in the OS keychain rather than a vendor cloud.
- Sandboxing is OS-native — sandbox-exec on macOS or bwrap on Linux — with network off unless approved per call.
- It exposes a REST API and MCP server/client modes, so automation requires connecting and routing through the dispatcher and its provider entries.
as of 2026-10-09
Verification history
We have re-verified Agenvoy 10 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-checked, vendor evidence unchanged
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 10 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Agenvoy tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Open Source
$0
Ideal for
Developers and power users on Linux or macOS who are happy to install, run and tune a local Go daemon rather than pay for managed hosting.
What this tier adds
Starting tier — AGPL-3.0 licensed, self-hosted as a single Go binary, with auto tool generation, MCP server and client modes, dispatcher routing across eleven provider entries, and commercial licensing available for closed-source use.
Where the pricing makes sense
The company stage and team size where Agenvoy's pricing actually pencils out — and where peers do it cheaper.
Agenvoy is AGPL-3.0 open source, with commercial licensing available for closed-source use, so cost is mostly your own provider API spend plus the machine and time to run the daemon. That puts it well below hosted agent platforms that bundle infrastructure and support into a subscription, and in the same bracket as other self-hosted harnesses where you supply the model endpoints yourself.
Setup time & first value
How long it actually takes to get something useful out of Agenvoy — broken out by persona, not the marketing-page minute.
Developers on Linux or macOS: roughly 10–20 minutes from the one-line curl install to a first tool build, then longer to wire in providers, skill packs and external MCP services. Team members only need an mcpServers entry to reach a daemon that is already running. Expect meaningful tuning time before cron jobs and subagent-based skills work the way you want.
Switching to or from Agenvoy
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a hosted agent platform: run the daemon locally and connect the same model providers through the dispatcher, then re-register your workflows as skill packs and cron schedules.
- →From Claude Code or Codex alone: add Agenvoy as an MCP server so your existing client keeps its session while gaining the shared sandboxed tool library.
- →From a self-rolled shell script cron setup: move each job to a skill pack bound to a cron entry so it hot-reloads on change and auto-disables on failure.
- →From a standalone RAG tool: register KuraDB as an MCP server and query it from the TUI, dashboard or Telegram instead of a separate app.
- ↗To a hosted agent platform: export what you can from SQLite under ~/.config/agenvoy, then rebuild schedules and tools in the vendor's cloud environment.
- ↗To Claude Code or Codex alone: keep calling any tools that have external dependencies directly and drop the daemon, accepting per-session capability resets.
- ↗To a plain cron plus shell scripts: rewrite the script and API tools as standalone files, losing sandboxing, snapshots and cross-session error memory.
Integrations
Resources & Guides
- Documentationagenvoy.com
Docs · Agenvoy
Full product docs from agenvoy.com
- Quickstartagenvoy.com
Quickstart · Agenvoy
Get up and running fast from agenvoy.com
- Documentationagenvoy.com
Rest Api · Agenvoy
Full product docs from agenvoy.com
- Documentationagenvoy.com
Built In Tools · Agenvoy
Full product docs from agenvoy.com
- Documentationagenvoy.com
Security · Agenvoy
Full product docs from agenvoy.com
- Documentationagenvoy.com
License · Agenvoy
Full product docs from agenvoy.com
Tutorials & Learning
YouTube returned 6 videos for “Agenvoy”, and we withheld 6: 6 could not be judged, because “Agenvoy” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Agenvoy.
Official links
Tools that pair well with Agenvoy
Common stack mates teams adopt alongside Agenvoy, with the specific reason each pairing earns its keep.
CowAgent
Open-source, self-hosted AI agent that plans tasks, runs tools, and grows its own long-term memory.
Fly.io
Fly.io runs AI agent sandboxes on hardware-isolated Linux VMs called Sprites that checkpoint automatically and bill by the second.
Zhipu GLM
Zhipu GLM is a full-stack large-model platform with open-weights GLM-5.3 coding models, MaaS APIs and autonomous agents.
Featured Head-to-Head Comparisons
Agenvoy vs Spider Cloud
If you need a self-hosted AI assistant that can autonomously write tools, schedule tasks, and interface with multiple LLMs, Agenvoy is unmatched. If your focus is extracting large-scale web data reliably for AI agents or RAG, Spider Cloud’s blazing fast Rust engine and new Browser AI commands make it the better pick. They’re complementary — use both together for maximum automation.
Agenvoy vs Presto Voice
For most developers seeking a free, self-hosted AI assistant with tool-building and MCP support, Agenvoy wins hands-down. Presto Voice is a specialized enterprise platform for QSR chains, not for individual users. Choose based on your domain: developer automation vs. drive-thru automation.
Agenvoy vs Temporal Ai
Choose Agenvoy if you're a developer wanting a self-hosted, always-on AI assistant that auto-creates tools and routes tasks across models. Choose Temporal if your priority is building fault-tolerant, durable workflows for AI agents and microservices that survive failures without losing progress. Agenvoy is a single-daemon tool-building agent; Temporal is a full workflow orchestration platform.
Alternatives to Agenvoy
View all
Frequently Asked Questions
Used Agenvoy? Help shape our editorial sentiment research.