# kagent

Bring AI agents to every cluster you run

Source: https://wheredoirunthis.com/products/kagent · JSON: https://wheredoirunthis.com/api/products/kagent · Read on 2026-10-08

## At a glance

- Type: Unknown
- Kind: Open source tool
- Isolation: Kata / gVisor
- Free tier: Not documented
- Starting price: Not disclosed
- GPU: No
- Regions: Not disclosed
- Coverage: 21 of 61 assessed capabilities supported
- Agent readiness: read 8 of 10 signals, drive 7 of 9 signals

## Links

- Homepage: https://kagent.dev/
- Docs: https://kagent.dev/docs/kagent/1.x/
- Repository: https://github.com/kagent-dev/kagent

## Runtime

- **GPU & accelerators**: No — "kagent does not provision GPU compute — it runs on YOUR Kubernetes cluster, so GPU is a property of your nodes. Local model serving (NVIDIA Nemotron via NemoClaw, Ollama, Hugging Face) would use your cluster's GPU nodes if present. Not a kagent-provided facet." (source: homepage + docs) (confidence: medium)
- **BYOC (Bring your own cloud)**: na — "na for a self-host OSS tool — kagent runs entirely inside YOUR own Kubernetes cluster/cloud by definition (covered by self-host-oss). BYOC as a managed-service 'deploy into your cloud account' concept does not apply." (source: self-host model) (confidence: high)
- **Secrets management**: Kubernetes Secrets stay outside agent environments; the egress gateway injects ModelConfig API keys and RemoteMCPServer secret-backed headers. — "Credential injection requires TLS termination and a provider for the reference authority. Missing prerequisites forward without credentials; attempted injection failure returns 403/503. Harness env has literal values, not secretRef support." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/networking-and-egress.md) (confidence: high)
- **Storage options**: PostgreSQL, optional pgvector agent memory, and operator-provided snapshot object storage. — "The kagent runtime memory feature uses the same PostgreSQL database; Harness snapshotPolicy.location specifies object storage." [source](https://kagent.dev/docs/kagent/1.x/agents/agent-memory.md) (confidence: high)
- **Supported workloads**: kubernetes-operations, multi-agent-workflows, coding-agents, standalone-sandboxes — "Harness runtimes: kagent, codex, claude, byo. Separate standalone sandboxes support process execution and file transfer. Old OpenClaw/Hermes AgentHarness was removed in 1.x." [source](https://kagent.dev/docs/kagent/1.x/agents/agent-harness.md) (confidence: high)
- **Persistence**: Postgres holds platform state and optional pgvector memory; idle Actors persist memory/filesystem snapshots in object storage. — "Sessions survive Worker release through snapshots; agent memory is separately configured on the kagent Harness runtime." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/suspend-and-resume.md) (confidence: high)
- **Data jurisdiction**: na — "na — data residency / region is entirely the operator's choice (wherever your cluster runs). No product-operated regions." (source: self-host model) (confidence: high)
- **Resource limits & quotas**: Operator WorkerPool capacity; standalone sandboxes default to 1 CPU/1Gi, TTL 1h and maxTTL 24h, configurable at installation. — "Activity does not extend sandbox lifetime; SandboxTemplates cannot override installation-wide CPU, memory or TTL policy." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/standalone-sandboxes.md) (confidence: high)
- **SDK languages**: go, python — "kagent runtime supports Go and Python engines. These are agent-building runtimes, not evidence of two platform-client SDK packages." [source](https://kagent.dev/docs/kagent/1.x/agents/agent-harness.md) (confidence: high)
- **Network & egress controls**: Default-deny per-Actor gateway policy derived from model/tool/skill/telemetry destinations; TLS termination enforces hostname rules. — "No matching rule or no policy denies outbound access. Policies cache for 10 seconds by default; an existing Session retains its pinned revision." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/networking-and-egress.md) (confidence: high)
- **Snapshots & forks**: Full/Data snapshots plus documented Session checkpoint and fork operations. — "Forking a checkpoint creates a second Session pinned to the captured revision. Full captures memory; Data preserves durable files and restarts. Standalone sandbox process handles do not survive suspend/resume." [source](https://kagent.dev/docs/kagent/1.x/examples/agents-via-mcp.md) (confidence: high)
- **Deployment methods**: helm, cli, dashboard, api — "Install with Helm; author Harness/AgentTemplate/Agent through Kubernetes and kagent APIs. CLI, web UI, gRPC and A2A surfaces are documented." [source](https://kagent.dev/docs/kagent/1.x/setup/installation.md) (confidence: high)
- **Isolation**: Kata / gVisor — "1.x runs every agent as an Actor in gVisor by default; kagent-generated ActorTemplates name gvisor-default. Registry groups gVisor under kata-gvisor-microvm: this does NOT mean kagent uses a microVM. Substrate supports a microvm class, but kagent-backed pools must stay on gvisor." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/sandboxing.md) (confidence: high)

## Capabilities

### AI & agents

| Capability | Status | Evidence |
| --- | --- | --- |
| Agent skills | Supported | "Skills provide agent knowledge/instructions; documented OCI-image and Git-source configuration." [source](https://kagent.dev/docs/kagent/1.x/skills-and-mcp/skills.md) |
| AI app builder | Not supported | "No prompt-to-app / vibe-coding generation surface." [source](https://kagent.dev/) |
| Autonomous QA agent | Not supported | "No first-party autonomous QA/testing agent (you could build one on kagent, but it is not shipped)." [source](https://kagent.dev/) |
| Built-in coding agent | Partially supported | "Built-in kagent agent runtime supports Go/Python engines and tools; Codex/Claude are separate third-party coding harnesses rather than a proprietary first-party coding agent." [source](https://kagent.dev/docs/kagent/1.x/agents/agent-harness.md) |
| BYOK for LLMs | Supported | "Bring OPENAI_API_KEY or other model-provider credentials; operator configures the Secret and namespace credential provider policy." [source](https://kagent.dev/docs/kagent/1.x/setup/installation.md) |
| Credential-injection proxy | Supported | "Kubernetes Secrets stay outside agent environments; the egress gateway injects ModelConfig API keys and RemoteMCPServer secret-backed headers. TLS termination and credential-provider configuration are required." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/networking-and-egress.md) |
| LLM gateway | Partially supported | "Documented bring-your-own agentgateway integration routes LLM traffic; the gateway is an external component, not a built-in hosted router." [source](https://kagent.dev/docs/kagent/1.x/setup/model-providers/byo-agentgateway.md) |
| MCP client | Supported | "kagent runtime agents consume RemoteMCPServer tools configured by AgentTemplates." [source](https://kagent.dev/docs/kagent/1.x/agents/agent-harness.md) |
| MCP-server hosting | Supported | "kagent integrates kmcp 0.4.0 and kagent-tools 0.3.0 to run tool servers in the operator cluster." [source](https://kagent.dev/docs/kagent/1.x/reference/versions.md) |
| MCP server | Supported | "The controller serves Model Context Protocol at /mcp on port 8083 over Streamable HTTP. Documented Session/checkpoint and sandbox tools; keep the unauthenticated default endpoint private." [source](https://kagent.dev/docs/kagent/1.x/examples/agents-via-mcp.md) |
| Model inference API (first-party) | Not supported | "Uses external/local model providers; no first-party model-serving catalogue established." [source](https://kagent.dev/docs/kagent/1.x/agents/agent-harness.md) |
| Preinstalled coding agent | Supported | "Harness runtime choices codex and claude run third-party coding agents in provisioned gVisor sandboxes with operator-supplied model credentials." [source](https://kagent.dev/docs/kagent/1.x/agents/agent-harness.md) |

### Networking & delivery

| Capability | Status | Evidence |
| --- | --- | --- |
| Automatic HTTPS | Not applicable | "Not a web host that provisions a public TLS endpoint per deployed workload; agent traffic runs over the service mesh (Istio/Ambient, mTLS)." [source](https://kagent.dev/) |
| Custom domains | Not applicable | "No user-facing hosted service with per-workload custom domains." [source](https://kagent.dev/) |
| Direct public IPv6 | Not applicable | "Not a raw-socket/VM product with per-instance public IPv6." [source](https://kagent.dev/) |
| Edge KV store | Not applicable | "Not an edge-runtime KV store." [source](https://kagent.dev/) |
| Egress controls | Supported | "Default-deny per-Actor gateway policy derived from model/tool/skill/telemetry destinations; TLS termination enforces hostname rules. Standalone Sandboxes have an empty policy and cannot configure outbound destinations." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/networking-and-egress.md) |
| Email send / receive | Not supported | "Agent channels cover Slack / Discord / Telegram (and A2A), not email send/receive." [source](https://kagent.dev/) |
| Public preview URLs | Not applicable | "Not a deploy-per-branch web host; no shareable per-deploy preview URL primitive." [source](https://kagent.dev/) |
| Private networking | Supported | "Agent networking runs inside the cluster through the Substrate gateway, with generated outbound policy. It is not a hosted private network product." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/networking-and-egress.md) |
| Built-in reverse proxy | Partially supported | "Controller can rewrite internal agent and MCP URLs through a configured gateway/proxy with x-kagent-host; not a general hosted reverse proxy." [source](https://kagent.dev/docs/kagent/1.x/operations/operational-considerations.md) |

### Organisation & security

| Capability | Status | Evidence |
| --- | --- | --- |
| Built-in auth | Not supported | "Default insecure mode admits every caller with self-selected identity. trusted-proxy does not verify token signature or expiry; an external validating boundary and network restrictions are required. Authorizer permits every check." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/identity.md) |
| BYOC | Not applicable | "Self-host OSS runs entirely inside your own cluster/cloud (covered by self-host-oss); BYOC as a managed-service 'deploy into your cloud account' concept does not apply." [source](https://kagent.dev/) |
| Deploy into customers' VPCs | Not applicable | "Self-host installation by the cluster operator, not a managed vendor control plane deploying into customer VPCs." [source](https://kagent.dev/docs/kagent/1.x/setup/installation.md) |
| Secrets management | Partially supported | "Operator-managed Kubernetes Secrets hold provider/tool credentials; gateway injects selected keys outside the agent. No standalone secrets vault." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/networking-and-egress.md) |
| Self-host / OSS core | Supported | "Apache-2.0 repository; install Agent Substrate, identity material, PostgreSQL and kagent into your Kubernetes cluster. No first-party hosted tier identified." [source](https://github.com/kagent-dev/kagent) |
| Sharing | Supported | "Session owner can share a conversation with another account; shares are Session-scoped, not Sandbox access. Default authentication is insecure, so sharing is not a security guarantee." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/identity.md) |
| SSO / SAML | Partially supported | "Bundled oauth2-proxy can provide browser sign-in, but controller authentication is separate and requires trusted-proxy plus upstream validation. No first-party SAML guarantee established." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/identity.md) |
| Teams | Partially supported | "Namespaces and creator-scoped Session lists aid collaboration; default caller identity is self-selected and permissive authorization is not tenant isolation." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/identity.md) |
| Workload identity | Partially supported | "Substrate components authenticate with mTLS and JWT authorities; this is infrastructure identity, not evidence of automatic third-party cloud IAM role assumption by workloads." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/identity.md) |

### Developer access

| Capability | Status | Evidence |
| --- | --- | --- |
| CLI | Supported | "Official kagent CLI manages agents/sessions and standalone sandbox create/exec/upload/download/suspend/resume/delete." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/standalone-sandboxes.md) |
| E2B-compatible API | Not supported | "Native kagent CLI/gRPC/MCP sandbox API; no E2B API or SDK compatibility advertised." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/standalone-sandboxes.md) |
| Filesystem API | Supported | "Sandbox file upload/download over gRPC (64 MiB bound) and MCP (1 MiB base64 chunks with continuation offsets)." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/standalone-sandboxes.md) |
| Git deploy | Supported | "'Define agents as Kubernetes CRDs — versioned in Git, reviewed in PRs, rolled out with the same tools you already use' (GitOps/ArgoCD). Skills also load from Git repos at startup." [source](https://kagent.dev/) |
| HTTP data API | Not applicable | "Data-plane HTTP query API primitive — not applicable to an agent platform." [source](https://kagent.dev/) |
| OpenAPI spec | Partially supported | "Published CRD field reference contains schema-level type and validation information; no aggregate downloadable OpenAPI found at common paths or linked from the reference." [source](https://kagent.dev/docs/kagent/1.x/reference/api-ref.md) |
| Private container registry | Not supported | "No hosted private container/image registry; you use your own registry for agent/sandbox images." [source](https://kagent.dev/) |
| Public API | Supported | "Documented Kubernetes CRDs, controller gRPC services, A2A and MCP APIs. Self-hosted API, not a vendor-operated endpoint." [source](https://kagent.dev/docs/kagent/1.x/about/architecture/kagent.md) |
| SDK | Partially supported | "Go and Python kagent agent-building engines plus BYO runtimes are documented; not independently verified platform client SDK packages." [source](https://kagent.dev/docs/kagent/1.x/agents/agent-harness.md) |
| SSH access | Not supported | "Process execution and file transfer via CLI/gRPC/MCP, not an advertised SSH endpoint." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/standalone-sandboxes.md) |

### Compute & storage

| Capability | Status | Evidence |
| --- | --- | --- |
| Cloud-init / ignition config | Not applicable | "Not a VM/microVM launch product with cloud-init user-data; workloads are configured via CRDs, not boot metadata." [source](https://kagent.dev/) |
| Desktop / computer-use sandbox | Not applicable | "No virtual desktop / computer-use GUI sandbox." [source](https://kagent.dev/) |
| GPU compute | Not supported | "kagent provisions no GPU compute — it runs on your cluster; local model serving (Nemotron/Ollama) would use your own GPU nodes. Not a kagent-provided facet." [source](https://kagent.dev/) |
| Live-state (memory) snapshots | Supported | "Full mode captures process memory and filesystem; Data mode preserves durable data but restarts execution. Standalone Sandboxes do not retain process handles on resume." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/suspend-and-resume.md) |
| Multi-region | Partially supported | "'Bring AI agents to every cluster you run' — agents pick the cluster they run in; multi-cluster/region placement is the operator's choice across your Kubernetes footprint. Not a product-operated multi-region service." [source](https://kagent.dev/) |
| Object storage | Not applicable | "Consumes operator object storage for snapshots; does not provision application buckets." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/suspend-and-resume.md) |
| Persistent disk | Partially supported | "Durable /data survives Data snapshots; Full captures memory and root filesystem as well. This is snapshot-backed state, not an attachable block-volume service." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/suspend-and-resume.md) |
| Sandbox templates / custom images | Supported | "SandboxTemplate defines commands and prepared files, layered onto an operator-configured guest image digest. Limits and TTL remain installation-wide." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/standalone-sandboxes.md) |
| Scale to zero | Supported | "Idle Actors can suspend and release Worker resources while snapshots persist. The shared Kubernetes control plane and WorkerPool do not disappear or become free." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/suspend-and-resume.md) |
| Selectable isolation runtime | Not supported | "Substrate supports multiple sandbox classes, but kagent agent pools must use gvisor; not a user-selectable kagent isolation-runtime feature." [source](https://kagent.dev/docs/kagent/1.x/substrate-runtime/sandboxing.md) |
| Snapshots / forking | Supported | "MCP checkpoint tools capture a Session and fork it into a second conversation at the pinned revision; Full/Data snapshots have different memory semantics." [source](https://kagent.dev/docs/kagent/1.x/examples/agents-via-mcp.md) |

### Data

| Capability | Status | Evidence |
| --- | --- | --- |
| Database branching | Not applicable | "Data-layer (copy-on-write DB branch) primitive — does not apply to an agent platform." [source](https://kagent.dev/) |
| Ephemeral instant DB (no-signup) | Not applicable | "No-signup claimable-DB primitive — not applicable." [source](https://kagent.dev/) |
| First-party database | Not applicable | "Operator provides PostgreSQL and pgvector for platform state and memory; not a managed database product for applications." [source](https://kagent.dev/docs/kagent/1.x/agents/agent-memory.md) |
| Vector search | Supported | "kagent Harness memory uses pgvector cosine similarity over 768-dimensional memories. Requires external pgvector PostgreSQL; not supported by codex/claude/byo Harnesses." [source](https://kagent.dev/docs/kagent/1.x/agents/agent-memory.md) |

### Platform & operations

| Capability | Status | Evidence |
| --- | --- | --- |
| Managed CI/CD runners | Not applicable | "Not a CI/CD runner platform." [source](https://kagent.dev/) |
| Built-in observability | Supported | "'Observable by default — OpenTelemetry traces, Prometheus metrics, structured logs. See every prompt, every tool call, every token.' Integrates Grafana + Langfuse." [source](https://kagent.dev/) |
| Preview environments | Not applicable | "Not a PaaS — no full ephemeral per-PR environment primitive." [source](https://kagent.dev/) |
| Scheduled tasks | Not supported | "No first-class cron/scheduled-job primitive advertised; agents are invoked interactively / via A2A / channels." [source](https://kagent.dev/) |
| Project template marketplace | Partially supported | "Public Agent registry (kagent.dev/agents) + Tool registry (kagent.dev/tools) — a gallery of prebuilt, reusable agent/tool definitions to adopt and customise. A catalogue of components rather than full project starters." [source](https://kagent.dev/agents) |
| Durable workflows | Partially supported | "Sessions/checkpoints, A2A delegation and tool approval/elicitation support multi-step work. Not a general Temporal-style durable workflow engine." [source](https://kagent.dev/docs/kagent/1.x/agents/human-in-the-loop.md) |

## Agent readiness

### Read: 8 of 10 signals present

8/10 signals (5 full, 3 partial)

- Llms txt: pass — "/docs/llms.txt and /docs/kagent/1.x/llms.txt return substantive indexes linking versioned guides, APIs and per-page .md sources. Apex /llms.txt remains 404."
- Llms full txt: fail — "/llms-full.txt returns 404; indexes link individual markdown pages, not an advertised full-document corpus."
- Robots AI policy: partial — "robots.txt (200) uses the Cloudflare 'content-signals' framework (search/ai-input/ai-train) with no explicit yes/no on the AI signals = 'neither grants nor restricts' = default-allow; no AI-UA Disallow."
- JSON LD: fail — "No <script type=application/ld+json> block on the homepage."
- Sitemap: pass — "/sitemap.xml returns HTTP 200 with 23 loc entries."
- Meta quality: pass — "Descriptive <title> + full OG (title/description/image 2400x1260) + twitter card."
- Markdown negotiation: partial — "Versioned .md pages return clean text/markdown; partial credit for explicit URL variants, not a verified Accept-header negotiation mechanism."
- Content efficiency: partial — "Homepage HTML 208357 bytes; visible text (HTMLParser excluding script/style) 8652 UTF-8 bytes = 4.15%, rubric 1–5% band. Inline SVG/Next.js payload inflate HTML."
- Docs discoverable: pass — "Docs first-class in the header nav ('Docs' dropdown) + footer; DocSearch (Algolia) search present; reachable without JS via sitemap."
- Pricing machine readable: na — "No pricing surface — kagent is free Apache-2.0 OSS (self-host). The /enterprise page points to Solo.io's separate commercial distribution, not kagent pricing tiers. Dropped from the denominator."
- Anti bot: pass — "Homepage, current versioned docs, robots and sitemap fetched without a challenge on 2026-10-08. Old unversioned docs return 500; recorded as a URL-migration issue, not a bot block."

### Drive: 7 of 9 signals present

7/9 signals (4 full, 3 partial)

- Public API: pass — "Documented controller gRPC, Kubernetes v1alpha3 resources, A2A and MCP; self-hosted endpoints, not a hosted service."
- Openapi spec: partial — "CRD reference has field and validation schemas; no aggregate downloadable OpenAPI found in that reference or common paths."
- MCP server: pass — "Documented controller Streamable HTTP /mcp on port 8083; concrete Claude Code/Cursor configs and curl examples for Session/checkpoint/sandbox tools."
- SDK packages: fail — "Go/Python agent engines do not by themselves verify official platform-client SDK packages. Prior full credit for runtime language count was unsupported."
- Auth docs: pass — "substrate-runtime/identity.md explicitly documents insecure and trusted-proxy modes, lack of signature/expiry verification and permissive authorization. Documentation credit is NOT security approval."
- Agent self serve creds: na — "Self-hosted OSS — there is no hosted signup / credential-issuance surface; you install it and supply your own LLM keys. Dropped from the denominator (needs a hosted endpoint)."
- Webhooks: fail — "Messaging-channel event integration (Slack/Discord/Telegram) exists but no documented webhook endpoint incl. signing was found."
- CLI tool: pass — "Official kagent binary and command reference; sandbox guide demonstrates install and lifecycle/file/process commands."
- Error model: partial — "MCP examples document JSON-RPC Invalid params (-32602), mismatched elicitation keys and polling transitions; no complete platform-wide error schema verified."
- Rate limit docs: na — "No vendor-hosted API or vendor rate-limit headers; self-hosted installation capacity and sandbox TTL limits are documented separately."
- Agent config: partial — "Documented skills and agent configuration. No additional AGENTS.md or well-known-agent surface verified in this capture."

## Coding harnesses

- Codex CLI (OpenAI) — "1.x Harness codex runtime runs the Codex coding agent." [source](https://kagent.dev/docs/kagent/1.x/agents/agent-harness.md)
- Claude Code (Anthropic) — "1.x Harness claude runtime runs the Claude coding agent." [source](https://kagent.dev/docs/kagent/1.x/agents/agent-harness.md)
- Cursor (Anysphere) — "Documented Cursor MCP settings connect to the controller /mcp endpoint; client integration, not a Cursor runtime backend." [source](https://kagent.dev/docs/kagent/1.x/examples/agents-via-mcp.md)

## Open source

- Repository: https://github.com/kagent-dev/kagent
- License: Apache-2.0
- Stars: 3965
- Forks: 825
- Self-host methods: helm, kubernetes, binary
- Self-host dependencies: kubernetes, postgres

## Company & presence

- Founded: 2025 (kagent created at Solo.io; CNCF Sandbox)
- Team size: Community project — maintained by Solo.io + contributors from Google, Confluent, Adobe, Red Hat, IBM, Microsoft
- Terms of service: https://github.com/kagent-dev/kagent/blob/main/LICENSE
- GitHub: https://github.com/kagent-dev/kagent
- Discord: https://discord.gg/Fu3k65f2k3
- LinkedIn: https://linkedin.com/company/kagent/
- YouTube: https://www.youtube.com/@kagentproject
- Blog: https://kagent.dev/blog
- CNCF: https://www.cncf.io/

## Nearby in the atlas

- [Agent Sandbox](/products/agent-sandbox.md)
- [Agent Substrate](/products/agent-substrate.md)
- [Agent37](/products/agent37.md)
- [AX (Agent Executor)](/products/ax.md)
- [Beam](/products/beam.md)
- [Cerebrium](/products/cerebrium.md)

## Method

Facts come from the product's own public pages. Each claim carries its evidence: a quote and the page it came from. We count capabilities rather than grade products, and "Not assessed" means we have not checked, never that the product lacks it.
