Documentation
TandemApp is one address and one dashboard in front of any number of engines and cloud APIs: the gateway, the dashboard, the Chat tab, the conductor and the fleet of engine hosts. These pages are the whole documentation; each covers one part and names the files that implement it. To install, start with install.md; to put a first engine or cloud API behind it, backends.md.
Installing
| page |
what it covers |
| install.md |
installing TandemApp: which way for whom, from a checkout, what a new install runs (the gateway only; Chat and the conductor switched on from the page), the first page, the prefixes a reverse proxy sends to TandemApp (controller-routes.json), where state lives, upgrades, backups, uninstall |
| install-linux.md |
the one command on Linux (a Raspberry Pi too) and macOS, its options, chat's sandboxes, and the container |
| install-kubernetes.md |
Helm on Kubernetes: the release archive built reproducibly, the install, ingress and TLS, the values, Chat with sandbox pods, the MCP bridge sidecar, upgrades and rollback, troubleshooting |
| auth.md |
sign-in, accounts and roles, API tokens and their scopes, who may use the API from other devices, recovery |
Backends, the gateway and the dashboard
| page |
what it covers |
| backends.md |
engines and cloud APIs behind TandemApp: the Add backend form and its providers, the kinds, vLLM, SGLang, tandem-app launch, any OpenAI-compatible server, OpenAI, Anthropic, DeepSeek, another TandemEngine, routing by model id, fallbacks, keys at rest, prices and cost, what is measured |
| proxy.md |
the gateway in depth: the Messages API and its translation, configuration, keys, the fleet, TandemNode, Add Spark, the data plane, the harness in front of a backend |
| monitoring.md |
several engines at once: every request in flight, vLLM's and SGLang's metrics pages (L1), their plugins (L2), the machine behind each backend, your own Prometheus |
| dashboard.md |
the dashboard view by view, the Live activity panel, the lanes, a request held for memory, the log export, the contract, access |
| chat.md |
the Chat tab (in beta) and chatd, the service behind it: who starts it, the sandbox, the port preview, the permission modes, Agents mode, the known limitations |
| settings.md |
one settings interface over the stores: scopes, change sets, versions and restore, engine restarts, audit |
The conductor
| page |
what it covers |
| conductor.md |
the conductor in one page: modes and headers, profiles, the router and its classes, effort, packages, the note, side requests, sessions, the journal, measured effect |
| session-monitor.md |
the session monitor: modes and policy, safe points and actions, every task, reasoning loops, the tool-call checks, the claim check, turn outcomes, session tasks, extending it |
| status-lines.md |
status lines in the client, the narrator and the voice, the shadow reader |
| chat-mode.md |
the conductor inside a TandemEngine: the TandemLLM model, the package library, Laya for the model's scripts, the session monitor's rules |
| packages.md |
the packages the conductor attaches by class: their shape, the first packages, how each is measured |
| memory.md |
memory notes: what the model keeps about the user and the work, how it writes and reads them |
Extending it
| page |
what it covers |
| hooks.md |
hooks: your code at each stage of a request (before routing, before the engine, on the answer, on an error), their order and limits, payload levels, signed webhooks in Python and Node, shadow mode |
| plugins.md |
plugins: write one (a folder with plugin.json and an MCP server), the manifest, capabilities, install, trust, the protocol, contributions, sandboxed plugins, debugging |
| plugins-sdk.md |
the plugin SDK: a plugin in Python as decorated tools, hooks, decision providers and backends, tandemapp-plugin new, the two examples built on it |
| decision-providers.md |
decision providers: your own model in place of the conductor's decider, per question; the systemone wire, shadow and A/B, health and fallbacks, the agreement report |
Reference
| page |
what it covers |
| reference/commands.md |
every flag of tandem-app serve, launch and hostcollect |
| reference/variables.md |
every TANDEM_* variable, the QSE_* variables TandemApp reads, the QWEN38_* knobs of the conductor inside an engine |
| reference/http.md |
every route, who may call it, every header and every error |
| reference/metrics.md |
every Prometheus family, and the usage ledger's columns |
| troubleshooting.md |
symptoms, causes and what to do: the dashboard, the Chat tab, vLLM and SGLang, the gateway |
| glossary.md |
TandemApp's terms |
| threat-model.md |
where untrusted input enters TandemApp, the parts that matter most for security |
| CHANGELOG.md |
every release |
tools/gen_reference.py writes the reference pages' tables from the code, between marked blocks. python3 tools/gen_reference.py --check, which tests/test_docs_reference.py runs, fails while a table is stale or the HTTP reference lacks a route or header the code has.
The API contracts are under ../contracts: the engine API (engine-api-v1), the engine extension interface (engine-ext-v1), the fleet (fleet-v1), the dashboard (dashboard-v1), the conductor (conductor-v1), the hooks (hooks-v1), the settings interface (settings-v1) and the plugins (plugin-v1). The engine has its own documentation in its repository (TandemEngine); on these pages a path under server/, engine/ or models/ names a file of the engine's repository.
The pages in this folder are CC BY 4.0 (LICENSE).