Agentfm Core is live · free · last verified 31 Oct 2026 · viability monitor
Agentfm Core
Run AI agents across a peer-to-peer mesh of the idle machines you already own
AgentFM Core is the rare decentralized-compute project that treats trust as a first-class feature rather than a slide: Ed25519-signed ratings, a per-peer Merkle log with RFC 6962 inclusion proofs, EigenTrust weighting so a rater with no history cannot tank a good worker, and witness nodes that refuse double-signers mesh-wide. The sandbox story is equally concrete — a throwaway Podman container per task, stdout streamed token by token over an encrypted libp2p channel, artifacts returned as a zip by task id. The catch is operational, not technical: you supply the machines, the container images, the relay and the uptime.
Verdict generatedVerified 4d ago · liveness 55/100 · cite: rightaichoice.com/tools/agentfm-core
- Developers with spare workstations or gaming rigs who want agents on owned hardware
- Researchers who need compute across machines they already run
- Small teams exploring peer-to-peer infrastructure for agent workflows
- Organizations routing internal agent traffic through a private encrypted mesh
- Teams that need an uptime SLA or a vendor support contract
- Buyers who want a turnkey hosted AI platform with no node management
- Anyone unwilling to run a binary with mode flags and manage Podman container sandboxing
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 AgentFM Core if you need a managed endpoint with an uptime SLA, or you are not prepared to run the agentfm binary with mode flags, build container images and keep a relay on a public IP.
You pay in machines and electricity rather than a subscription: the compute is yours, so idle-node uptime and power draw become your operating cost.
AgentFM Core is Apache-2.0 and self-hosted, so the cost comparison is against hardware you already own versus per-hour GPU rental. A single spare gaming rig running overnight can absorb batch agent work that would otherwise be metered by the hour on a cloud endpoint; the trade is that you carry the relay, the container images and the uptime yourself.
In short
Agentfm Core — Run AI agents across a peer-to-peer mesh of the idle machines you already own. Best for Developers with spare workstations or gaming rigs who want agents on owned hardware, Researchers who need compute across machines they already run, Small teams exploring peer-to-peer infrastructure for agent workflows. Free to use.
What is Agentfm Core?
Agentfm Core — Run AI agents across a peer-to-peer mesh of the idle machines you already own. Best for Developers with spare workstations or gaming rigs who want agents on owned hardware, Researchers who need compute across machines they already run, Small teams exploring peer-to-peer infrastructure for agent workflows. Free to use.
AgentFM Core turns spare CPUs and GPUs — a workstation, a gaming rig, an office mini PC — into a private compute mesh for AI agents. You package an agent as a container, install the agentfm binary on each machine, and dispatch tasks from the macOS desktop app or any OpenAI-compatible client. Every node runs the same binary and a flag picks the role: -mode worker runs your agent in a fresh Podman sandbox with no host access and streams stdout back token by token; -mode relay handles discovery, NAT traversal and ledger archival on a public IP; -mode api exposes a headless HTTP and OpenAI-compatible gateway so existing SDK clients point at the mesh without rewrites. Workers broadcast live CPU, GPU, RAM and queue depth every two seconds, and dispatch routes to the least-loaded node advertising the right model. Trust is explicit rather than assumed: every rating is Ed25519-signed and appended to an append-only per-peer Merkle log with RFC 6962 inclusion proofs, EigenTrust weighs raters by their own standing, and witness nodes eject double-signers mesh-wide. Files an agent writes come back as a zip keyed to the task id. It is Apache-2.0, written in Go on libp2p, and aimed at developers, researchers and small teams who would rather schedule work across hardware they control than rent cloud GPUs by the hour. You supply the machines, the container images, the relay and the uptime — this is not a managed endpoint with a support contract.
Key features
- Peer-to-peer mesh network for AI agent execution over libp2p
- Worker mode runs each agent task in a fresh Podman sandbox with no host access
- Relay mode with DHT discovery, NAT traversal and ledger archival
- OpenAI-compatible HTTP gateway via -mode api on a configurable port
- Python SDK: pip install agentfm-sdk (AgentFMClient)
- End-to-end encrypted libp2p task streams between chosen machines
- Live token-by-token stdout streaming during dispatch
- Agent-written files returned as zipped artifacts keyed to task id
- Ed25519-signed ratings with RFC 6962 Merkle inclusion proofs
- EigenTrust reputation weighting that gates dispatch
- Witness nodes that catch and refuse double-signers mesh-wide
- Live node telemetry: CPU, GPU, RAM and queue depth every 2 seconds
- Private swarms via swarm key (-mode genkey)
- Interactive terminal dispatcher (-mode boss)
- Load circuit breakers via -maxcpu, -maxgpu and -maxtasks flags
Behind the verdict
AgentFM Core is best understood as infrastructure you operate, not a service you subscribe to. The unit of work is a container image: you write a Dockerfile or Containerfile, point -agentdir at it, pass -image, and the worker runs each task in a throwaway Podman sandbox with no host access, an injected .env and an attached GPU. That is a genuinely clean execution model — every dispatch gets a fresh container rather than a warm process that carries state between jobs, and the host filesystem is out of reach by design. Distribution is where the project puts its engineering. One binary, five role flags. -mode worker joins the mesh and advertises a model, capability tag and queue slots; -mode relay sits on a public IP and handles DHT discovery, NAT traversal and ledger archival so trust history survives restarts; -mode api puts an OpenAI-compatible HTTP gateway on a configurable port so an existing SDK client points at your own hardware without a rewrite; -mode boss is an interactive terminal dispatcher; -mode genkey generates a private swarm key. You can see the whole flow in the shipped example: curl /api/workers, copy a peer id, POST /api/execute with a prompt, and stdout streams back with curl -N. The trust layer is the part competitors tend to skip. The verifying key is derived from the rater's peer id, so you cannot forge or edit another peer's rating; entries land on an append-only per-peer Merkle log with RFC 6962 inclusion proofs and tampering breaks the root; EigenTrust weights each rater by their own standing; witnesses catch equivocators and refuse them work across the mesh. Load circuit breakers (-maxcpu, -maxgpu, -maxtasks) keep a worker from being buried. What you take on: Podman and container images are part of setup, a relay on a public IP is required for NAT traversal and ledger archival, and throughput tracks your spare capacity — when the machines are busy, the queue waits. A brand-new peer starts with no standing, so reputation-gated dispatch is cold at first. The macOS beta is not notarized with Apple, so first launch is blocked until you approve it via System Settings, Privacy & Security, Open Anyway, or clear the flag with xattr -d com.apple.quarantine. The desktop app (M-series and Intel) does bundle a full node, which is the gentlest entry point. Teams that need an uptime SLA or vendor audit paperwork should look elsewhere; teams with idle hardware and a willingness to run a binary with flags get encrypted, self-hosted agent execution with verifiable receipts.
Models under the hood
as of 2026-11-01
Limitations
- Agents run on hardware you own, so throughput tracks your spare capacity rather than a vendor quota — when your machines are busy, the queue waits.
- The macOS beta is not notarized with Apple, so first launch is blocked until you approve it via System Settings, Privacy & Security, Open Anyway, or clear the flag with xattr -d com.apple.quarantine.
- Sandboxing means Podman and container images (a Dockerfile/Containerfile with -agentdir and -image) are part of your setup.
- The mesh relies on a relay on a public IP for NAT traversal and ledger archival, and dispatch is reputation-gated, so a brand-new peer starts without standing.
as of 2026-10-31
Researching Agentfm Core? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is Agentfm Core still active?
Agentfm Core is live · free · last verified 31 Oct 2026 · viability monitor.
Viability score
How well maintained and how widely used is Agentfm Core? 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: November 2026
How we score →What do users say about Agentfm Core?
What people actually say about Agentfm Core — is it worth it?
Our community sentiment research written Jul 3, 2026 is more than 90 days old, so it is being re-reviewed against current data and is not shown.
How we verified this
Every listed fact about Agentfm Core — pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 31 Oct 2026. Analysis is drafted with AI models and checked against our editorial standards before publishing; see how we verify.
Verification history
We have re-verified Agentfm Core 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.
- — changed integrations, features, who should skip it, hidden costs, setup time, pricing fit, switching out, our verdict, description, switching in, limitations
- — changed use cases, description, our verdict, features, hidden costs, limitations, switching in, switching out, models, who should skip it, pricing fit, setup time, summary, who it suits
- — changed description, who should skip it, our verdict, summary, who it suits, features
- — changed platforms
- — changed SSL certificate
- — changed domain expiry
Showing the 6 most recent of 10 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does Agentfm Core cost in 2026?
Agentfm Core: Free to use. Pricing last verified 31 Oct 2026.
Where the pricing makes sense
The company stage and team size where Agentfm Core's pricing actually pencils out — and where peers do it cheaper.
AgentFM Core is Apache-2.0 and self-hosted, so the cost comparison is against hardware you already own versus per-hour GPU rental. A single spare gaming rig running overnight can absorb batch agent work that would otherwise be metered by the hour on a cloud endpoint; the trade is that you carry the relay, the container images and the uptime yourself.
Has Agentfm Core's pricing changed?
No plan price for Agentfm Core has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick Agentfm Core?
Pick Agentfm Core if you are Developers with spare workstations or gaming rigs who want agents on owned hardware, Researchers who need compute across machines they already run, Small teams exploring peer-to-peer infrastructure for agent workflows, Organizations routing internal agent traffic through a private encrypted mesh.
Look elsewhere if you are Teams that need an uptime SLA or a vendor support contract, Buyers who want a turnkey hosted AI platform with no node management, Anyone unwilling to run a binary with mode flags and manage Podman container sandboxing, Organizations whose compliance process requires vendor audit paperwork.
Skip it if Skip AgentFM Core if you need a managed endpoint with an uptime SLA, or you are not prepared to run the agentfm binary with mode flags, build container images and keep a relay on a public IP.
GeneratedReal-world workflow fit
Concrete scenarios for the personas Agentfm Core actually fits — and what changes day-one when you adopt it.
Install with the curl one-liner on both machines, start the macOS desktop app, then run the gaming rig with -mode worker -agentdir ./my-agent -image ghcr.io/you/myagent:v1 -model llama3.2 -maxtasks 4 and dispatch a document-drafting task from the app.
Outcome: Output streams back token by token over an encrypted libp2p channel, the written report returns as a zip on the artifact protocol, and the rating you give is Ed25519-signed into the Merkle log.
Stand up -mode relay on a public-IP box for discovery, NAT traversal and ledger archival, run workers with -mode worker and -capability tags, and point existing OpenAI SDK clients at -mode api -apiport 8080.
Outcome: Existing SDK clients dispatch to the least-loaded worker that advertises the right model without code rewrites, and nothing leaves the machines you chose.
Run competing workers with different standing, generate a private swarm with -mode genkey, and inspect the per-peer logs with RFC 6962 inclusion proofs after each dispatch.
Outcome: You can observe EigenTrust weighting in practice and see witness nodes refuse a double-signer mesh-wide, with signed receipts to verify the chain.
Use cases
- Run agent workloads across your own workstations instead of renting cloud GPUs by the hour
- Route internal agent traffic through a private encrypted mesh that never leaves your machines
- Contribute a gaming rig or office mini PC as worker capacity overnight
- Prototype peer-to-peer task distribution with verifiable signed results for research
- Point an existing OpenAI SDK client at -mode api with no code rewrite
- Dispatch agent jobs from the macOS desktop app with a trust badge on every agent
- Collect every file your agents produce as zipped artifacts organised by project
- Test reputation-gated scheduling by running competing workers with different trust scores
Setup time & first value
How long it actually takes to get something useful out of Agentfm Core — broken out by persona, not the marketing-page minute.
If you already have Podman and a container image, the macOS desktop app bundles a full node and gets you to a first local dispatch in roughly 15-30 minutes. Adding a second machine as a worker is another 10-15 minutes per host; standing up a relay on a public IP and getting NAT traversal working is the step that usually eats an afternoon.
Switching to or from Agentfm Core
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From cloud GPU endpoints: point existing OpenAI SDK clients at -mode api -apiport 8080 instead of the hosted base URL, then move batch work onto worker nodes.
- →From a single-machine agent runner: package the agent as a Dockerfile, set -agentdir and -image, and let -mode worker sandbox each task in a fresh Podman container.
- →From ad-hoc ssh scripts: replace them with -mode boss for interactive dispatch or the Python AgentFMClient for programmatic task runs.
- ↗To a managed agent endpoint: the -mode api gateway already speaks OpenAI, so clients can be repointed back to a hosted base URL with minimal changes.
- ↗To a container scheduler: agent images are plain container images, so the same Dockerfile and -image you pass to -mode worker can run under your existing orchestration.
Integrations
Resources & guides
Tutorials & Learning
YouTube returned 6 videos for “Agentfm Core”, and we withheld 6: 6 did not mention Agentfm Core. We are showing none, because we could not prove any of them are about Agentfm Core.
Official links
What are the best Agentfm Core alternatives?
The closest alternatives to Agentfm Core we list are Mesh Llm, Cerebras and MAX Engine. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| Mesh Llm | Freemium | Pricing not published | — |
| Cerebras | Freemium | Pricing not published | — |
| MAX Engine | Freemium | Pricing not published | — |
Alternatives to Agentfm Core
View all
Mesh Llm
Mesh LLM splits large open-weight models into layer ranges across your own GPUs and serves them through one OpenAI-compatible endpoint at localhost:9337.
Cerebras
Cerebras inference cloud: frontier LLMs at up to 750 tokens/sec for real-time agents and voice AI.
MAX Engine
MAX Engine serves open models through one OpenAI-compatible endpoint across NVIDIA, AMD, Apple silicon, CPUs, and ASICs.
Tools that pair well with Agentfm Core
Common stack mates teams adopt alongside Agentfm Core, with the specific reason each pairing earns its keep.
Mesh Llm
Mesh LLM splits large open-weight models into layer ranges across your own GPUs and serves them through one OpenAI-compatible endpoint at localhost:9337.
Cerebras
Cerebras inference cloud: frontier LLMs at up to 750 tokens/sec for real-time agents and voice AI.
MAX Engine
MAX Engine serves open models through one OpenAI-compatible endpoint across NVIDIA, AMD, Apple silicon, CPUs, and ASICs.
Agentfm Core vs its closest rival
Agentfm Core's closest head-to-head is with Spider Cloud.
Our Agentfm Core vs Spider Cloud verdict written Jul 3, 2026 is more than 90 days old, so it is being re-reviewed against current data and is not shown.
Frequently asked questions
Categories
Best-of guides
For makers of Agentfm Core
- Get a Verified Listing for Agentfm Core ($39): a written decision within 1 working day.
- Free listing for another tool: Free reviews are decided in order, within 3 weeks.
Used Agentfm Core? Help shape our editorial sentiment research.