Build with Platos

Durable, versioned agents with their own prompt, tools, models, memory, and budgets.

Version every change to an agent and roll back instantly without losing conversation history.

Group agents that share memory, threads, and user identity so a chat agent and a background agent feel like one.

Reusable agent behaviors with manifests, required env vars, and Claude-skills compatibility.

First-party skills shipped with Platos including code execution, csv-ops, and platos-rag.

How Platos routes a tool call from the agent to a connected entity, a skill, or a meta-tool.

External backends that connect to Platos over WebSocket and expose their tools to agents without exposing their data.

Long-term, scoped, ratable memory that powers personalization without leaking across tenants.

Knowledge-graph view over memory, with entity nodes and relations extracted from conversations.

How a conversation is stored, encrypted, branched, forked, and threaded.

The agent chat playground and the Postman-style request console for inspecting tool pipelines.

The four-tier context resolver that injects user, session, agent, and project values into tool calls and prompts.

How Platos assembles a prompt, when it caches, and how prompt blocks compose.

Pick a model per agent, compare two models head to head, and see which providers must be linked.

Bring your own provider keys (OpenAI, Anthropic, etc.) and link them per environment with a checklist.

Define eval criteria once, run them against golden sets, and see ratings roll up per agent.

Compare two agent versions head to head on the same set of inputs.

Platos-generated structured outputs (markdown, code, html-sandboxed, json, csv, svg, image) that render consistently in any client.

Upload files, images, and PDFs into a conversation and let the agent reference them as multimodal inputs.

Long-running background operations spawned from agents via the spawn_bgo meta-tool.

Execution records of trigger.dev tasks. The engine-layer primitive behind every BGO and tool block.

Cron-style and one-shot scheduled trigger.dev tasks. Used by Platos for cost reconcile, approvals expiry, and memory extraction.

trigger.dev queues that gate task concurrency. Each Platos environment has at least one queue per task type.

Code-level deployments of trigger.dev task bundles. The engine equivalent of an agent version, but for the task code itself.

Group of trigger.dev runs spawned together. Used by Platos for the agent_batch meta-tool.

Durable wait-for-token primitive that lets a run pause for hours or days awaiting human or external input.

Pause a tool call until a human approves or rejects, with durable waitpoints that survive restarts.

Per-agent and per-environment spend caps that throttle or block turns when the budget is exceeded.

Per-agent PII filters, safety event logging, and the governance dashboard that rolls them up.

Per-IP, per-token, per-agent, and per-environment rate limits enforced at the agent runtime.

Where keys live, how conversations are encrypted at rest, and the single secret-store invariant.

Per-agent and per-user monitoring dashboards with cost, request volume, and ratings.

OpenTelemetry-style trace view of a single turn, with prompt + tool spans + cost samples.

How Platos counts every input + output token, plus shadow spend on embeddings, extraction, and judge calls.

Prometheus-style metrics endpoint and the custom dashboards that consume them.

Append-only audit trail of admin actions and tool calls, scoped per environment.

Common questions about Platos — what it is, how it compares, where to deploy, how to use it.

What play.platos.dev is, what it isn't, and where to deploy your real agents.

The (organizationId, projectId, environmentId) tuple that gates every scoped row in Platos.

Three auth modes (direct headers, session-token JWT, service secret on WS) plus user-token passthrough.

First-party client libraries for Platos, including the consumer @platosdev/client and the entity-side platools SDKs.

Drop-in React FAB chat widget — three identity flows (anonymous form, OTP-verified, backend-authenticated), every per-turn agent option exposed, fully themable.

Universal MCP endpoint that federates entity tools, trigger meta-tools, Platos skills, and the control plane.

Personal Access Tokens (pmt_*) and OAuth 2.1 DCR flows for MCP clients.

The HTTP REST surface auto-generated from controllers, with a downloadable OpenAPI spec.

Outbound HTTP webhooks for conversation lifecycle events, run state transitions, and safety hits.

Share an agent publicly via a guest-token route, or drop in a web-component embed.

Run Platos on your own infra with a single docker compose up. Postgres, Redis, ClickHouse, MinIO, agent, webapp.

Per-project dev/staging/prod environments with their own scoped env vars, providers, and queues.

Real-time stream of token deltas, tool calls, artifacts, and run updates over WebSocket.

Where to find Platos's legal documents, security disclosure process, and what each covers.