llm-router
v1.5.5One front door + provider protocol in front of every LLM provider.
Install
iii trigger compose::add worker=llm-routerRun in your project with the iii engine running. Exact versions are immutable and digest-pinned.
llm-router
One front door for every LLM provider. The router owns routing, the provider registry, credential resolution, the model catalog, streaming relay, retries, and a single failure contract — consumers call one chat surface and never talk to a provider directly.
llm-router is a standalone iii worker. Providers plug in as separate workers
at runtime through a self-registration protocol
(iii call compose::add worker=provider-<x>); the router never compiles against a provider, and removing a
provider worker removes the provider.
Install
iii call compose::add worker=llm-routerQuickstart
A consumer streams a turn by creating an iii channel, handing the router the
channel's write endpoint, and reading frames from the read endpoint
while router::chat runs. Any SDK works; Node shown:
import { createChannel } from 'iii-sdk/helpers';
const { reader, writerRef } = await createChannel(iii);
reader.onMessage((frame) => {
const event = JSON.parse(frame); // AssistantMessageEvent
if (event.type === 'text_delta') process.stdout.write(event.delta);
});
const res = await iii.call({
function_id: 'router::chat',
payload: {
writer_ref: writerRef, // direction "write"
model: 'claude-sonnet-4',
messages: [{ role: 'user', content: [{ type: 'text', text: 'Hello' }], timestamp: Date.now() }],
},
timeoutMs: 620_000, // outer timeout ≥ the router's 600s stream budget
});
// res: { ok, provider, model, stop_reason, usage }The streaming contract: every stream ends with exactly one terminal frame
(done or error). When the router has to kill a stream itself (idle
timeout, provider crash), it synthesizes the terminal frame and attaches the
partial content, so consumers never hang on a half-open stream.
Both router::chat and router::complete accept an optional retry_max:
the maximum number of retries after the first provider attempt. Omit it
to keep the router's configured retry behavior, or pass 0 when the caller
must authorize and budget each attempt itself. A caller can only reduce the
configured limit: the effective ceiling is the smaller of the two, fixed
for the turn. This bounds router dispatches; it does not change a provider's
own retry policy or make missing usage a billable cost of zero. If a caller
may need to abort, it must send its own unique request_id with the original
request: an omitted id is generated internally and is not returned in the
response. Caller timeouts still require explicit router::abort with that
same caller-supplied request_id.
Functions
Consumer surface
| Function | Purpose |
|---|---|
router::chat |
Stream a turn into the caller's channel; returns the turn summary. |
router::complete |
Non-streaming convenience over the same pipeline; returns the final message. |
router::abort |
Cancel an in-flight turn by request_id. |
router::route |
Read-only routing preview: {model, provider?} → {provider, candidates}, same rules and error codes as router::chat. Pin the result as the explicit provider on the chat call when you need the provider before streaming. |
router::models::list |
List catalog models, filterable by provider / capability / modality (chat by default, stt, tts, any). |
router::models::get |
Fetch one model record (null when unknown). |
router::models::supports |
Check one capability flag for one model. |
router::embed |
Batch text embeddings through the first embed-capable provider in the registry, then embed-only workers ({model?, provider?, input[]} → {provider, model, embeddings[][]}); providers without an embed surface are skipped. |
router::transcribe |
Speech to text: {model?, provider?, audio_base64, mime?, language?, prompt?} → {provider, model, text, segments[]?, language?, duration_secs?}. The provider comes from the named provider, else from the catalog owner of the stt model, else the first provider that declared one. |
router::speak |
Text to speech: {model?, provider?, text, voice?, format?, language?, speed?} → {provider, model, audio_base64, mime, voice?, duration_secs?}. Same resolution over tts models. |
router::count_tokens |
Count prompt tokens: {model, provider?, system_prompt?, tools?, messages, thinking_level?, response_format?} → {provider, model, tokens, estimator}, resolved with the same routing rules as router::chat and forwarded to provider::<id>::count_tokens. Never runs the model and costs nothing; estimator is provider (metering API) or tiktoken (local tokenizer). A provider without the surface is a typed router/no_token_counter error, so callers can fall back to their own estimate. |
router::provider::list |
Registered providers with configured / available status, where each credential comes from (credential_source, credential_ref, credential_error — never a value), and each provider's starting point: default_model and default_thinking_level (see Default model). |
Only the read surface is agent-callable (router::models::list / get /
supports, router::provider::list); everything else is denied to in-run
agents — see Security model.
Error conventions
Errors use stable router/<code> identifiers. Streaming failures return an
error object with a user-facing message, the shared kind, retryable,
and an optional diagnostic detail. User interfaces should show message
and keep code / detail behind technical details; automation should branch
on code and retryable, never parse the prose.
The main recovery codes are router/provider_unavailable,
router/capacity_exceeded, router/stream_idle_timeout,
router/stream_incomplete, router/provider_auth_expired,
router/provider_rate_limited, router/context_overflow, and
router/provider_rejected.
Provider protocol
Token-gated after the first declare: the response to register carries a
registration token, and every later protocol call must present it.
| Function | Purpose |
|---|---|
router::provider::register |
Self-declaration at attach time; idempotent re-declare with the token. |
router::provider::resolve |
Per-request credential + endpoint resolution (config > secret:// / env:// reference > env > none, see Credentials). |
router::provider::update_credential |
Persist a refreshed credential (OAuth write-back). |
router::models::reconcile |
Replace the provider's catalog slice in one write. |
These four are registered with observable: false: their payloads carry the
registration token (every one) or a credential (resolve's output,
update_credential's input), so an observe of any of them fails with
NOT_OBSERVABLE.
The provider worker itself exposes provider::<id>::stream and, when
capable, provider::<id>::refresh_models (model discovery),
provider::<id>::count_tokens (prompt token counting),
provider::<id>::embed (embeddings), and the speech pair
provider::<id>::transcribe / provider::<id>::speak.
A worker that serves only embeddings registers provider::<id>::embed and
never calls router::provider::register, so it stays out of
router::provider::list and out of every model picker; iii-directory's local
MiniLM (minilm) is one. A router::embed call that names no provider tries
these embed-only surfaces last, alphabetically, found through
engine::functions::list in the router's namespace, and only when every
registered provider either has no embed surface (Anthropic only, say) or is
not configured for good (provider-openai with no API key). Configuring that
provider later moves callers to its vector space: memory re-embeds into the
model the reply reports. One rule blocks the fallback with
router/provider_unavailable, since another embedder would put the caller's
vectors in another space, so the caller retries later: a registered provider
whose embed has answered before (its registry record keeps an embeds flag)
is not found and was live in the last 15 minutes (its registration, its last
answered embed, or the first 3 minutes after the router starts), whatever else of it is live
(restarting, partly re-registered, or the router just restarted). After that
window it no longer blocks: the registry keeps every record, so a removed
provider must not hold the fallback for good.
router::chat accepts an optional structured prompt next to the flat
system_prompt: system_sections (ordered {text, cache_boundary} blocks)
and cache_intent ({surface_digest}, the caller's sha256: identity for
the text before the boundary). A sections-only request is flattened into
system_prompt (sections joined with \n\n); when both are sent they must
agree byte for byte or the request is rejected as invalid_request. Both are
forwarded verbatim to provider::<id>::stream alongside the flat string, so a
provider that knows nothing about sections keeps reading system_prompt.
Cache-aware providers put their prefix marker on the boundary block
(Anthropic, Claude Code) or derive a shared prompt_cache_key from the digest
(OpenAI, Codex) so every session on the same frozen prefix lands on one cache
entry. The digest is a routing hint, never evidence of a hit — read
usage.cache_read for that.
Configuration
All operator configuration lives in the engine's llm-router configuration
entry — no env vars, no config file. The entry schema is composed at runtime
from each registered provider's declaration. The router fetches the complete
entry at boot and keeps an in-memory snapshot synchronized by its observer of
configuration::on-updated (the entry's configuration:updated events);
request handlers never poll configuration::get.
{
"default_provider": "anthropic",
"providers": {
"anthropic": {
"api_key": "secret://ANTHROPIC_API_KEY",
"api_url": "https://api.anthropic.com/v1/messages",
"max_tokens": 8192
}
},
"routing_heuristics": [{ "pattern": "^gpt-", "provider": "openai" }],
"settings": {
"stream_timeout_ms": 600000,
"idle_timeout_ms": 120000,
"retry_max": 2,
"output_token_max": 32000
}
}| Setting | Default | Meaning |
|---|---|---|
stream_timeout_ms |
600000 |
Hard budget for one streamed turn. |
idle_timeout_ms |
120000 |
Max silence between provider frames before the attempt is cut. |
retry_max |
2 |
Retries per turn for retryable failures before the first forwarded frame (0–10). |
output_token_max |
unset | Optional ceiling on max_output_tokens. Unset forwards each model's own output ceiling, so providers run without an artificial limit; set it only to cap spend. |
Pasting a key into a provider's slice is the whole onboarding flow: the
router diffs the changed slice, debounces ~2 s, and kicks that provider's
provider::<id>::refresh_models discovery; discovered models land in the
catalog via router::models::reconcile and show up in router::models::list
within seconds — no restart.
Credentials and secret references
The configuration service's default filesystem adapter persists this entry at
./config/<entry-id>.yaml (llm-router unless III_CONFIG_NAME overrides it),
a folder meant to be committed, so keep keys out of it: keep the key in the
secrets worker, which llm-router depends on, and put a versionable
reference in the slice:
"api_key": "secret://ANTHROPIC_API_KEY"— encrypted in the secrets worker's vault;"api_key": "env://ANTHROPIC_API_KEY"— the variable in the project's env file (.env, or the secrets worker'senv_filefor this namespace; else the secrets worker's environment), read by the secrets worker each time, so an edit to the file applies without restarting the router.
Provider keys reuse their env var name; NAME matches
^[A-Za-z_][A-Za-z0-9_.-]{0,127}$. Either way the secret's consumers must
list llm-router; the Console's key field does this for you.
A provider's credential is the first of:
- the slice's
credentialobject (OAuth write-back viaupdate_credential); - a literal
api_key; - an
api_keyreference (secret://orenv://), resolved throughsecrets::resolve; - the provider's
credential_env_var, read in the router's process; - none.
A reference is never forwarded to a provider as a key. When it does not
resolve, the provider is not configured, and the error says why: secret NAME not found in the secrets worker, environment variable NAME is not set in the project's .env or the secrets worker's environment, llm-router is not allowed to read secret NAME; add llm-router to the secret's consumers, or
secrets worker is not running. The env var is deliberately not used as a fallback past an
explicit reference: you chose the stored secret, so a missing one is reported
rather than silently replaced by a different key. Providers built on the
shared scaffold skip their own env fallback in that case too, and
router::chat fails before dispatch with router/not_configured and the same
reason.
router::provider::list entries and the router::provider::resolve response
carry credential_source (config | env | secret | none),
credential_ref (the secret:// or env:// reference, when the slice uses
one; credential_source is secret for both, as the secrets worker resolved
it) and
credential_error. All three are optional on the wire. The older source
field keeps its config | env | none values: a resolved reference reports
config there.
Resolved values live only in the router's memory: never in the entry, logs or traces, and a cached value is wiped when replaced. They are re-read:
- on
secrets::on-changed(the router observes it at boot; without a secrets worker the observer waitspendinguntil the function registers). A rotated, deleted or revoked secret then nudges every provider whose slice references it throughprovider::<id>::on_router_ready, so the provider drops its cached resolve, re-declares and refreshes its models; - when a slice that references it changes, before the debounced
refresh_modelsfires, so a pasted reference is resolvable by the time discovery asks for it; - when the secrets worker registers after the router, so a router that booted first recovers without a restart;
- on demand after 5 minutes (10 seconds for a failure), bounding staleness if an event is ever missed. While the secrets worker is unreachable, a value already resolved keeps being served: secrets cannot rotate or be revoked while it is down.
The reference itself is not secret: it is committed, listed and logged. Never
paste a key after secret:// or env://.
Operational notes
- Secret references resolve as
llm-router. The secrets worker checks the router's engine-stamped worker identity against the secret'sconsumers, not the provider's. - Env-var credential fallback resolves in the router's process. Without
a reference, a provider's
credential_env_var(e.g.ANTHROPIC_API_KEY) is read by the llm-router binary, not by the provider worker — launch the router with those variables set, or put a reference in the entry. A key present only in another worker's environment shows up asconfigured: false.env://NAMEis different: the secrets worker reads it from the project's.env, so the router needs noenv_filefor it. - Registration-token recovery. Re-registering a provider id without its
original token is rejected (anti-takeover). If a provider durably lost its
token, delete the router's registry state (state scope
llm-router, keyregistry) and restart the affected providers to re-bind; pasted credentials in the configuration entry are unaffected.
Security model
The agent-callable surface is governed by the engine-wide permission policy
(repo-root iii-permissions.yaml), not a per-worker
file. In-run agents may only read the catalog and provider list:
- Allowed:
router::models::list,router::models::get,router::models::supports,router::provider::list. - Denied to agents: the chat/spend surface (
router::chat,router::complete,router::abort,router::route), the whole provider/credential protocol (router::provider::resolve/register/update_credential,router::models::reconcile), and directprovider::*calls.
Worker-to-worker calls bypass the agent gate, so the harness, context-manager,
and provider workers reach the full surface — only in-run agents are
restricted. Provider credentials live in the configuration entry, or behind a
secret:// or env:// reference in the secrets worker, and are never readable back
through any allowed function: router::provider::list reports where a
credential comes from, never its value. The provider protocol functions are
not observable, so registration tokens and credentials reach no observer (see
Provider protocol).
Events
The router hosts three event functions. Each is private (only the router
calls it) and returns the event it is called with, so its observers observe
the event; none takes args, so observers select events with filters on
input.
| Function | Called when | Event | Filter on |
|---|---|---|---|
router::models::on-changed |
a provider reconciles its catalog slice | { "provider": "<id>", "count": <n> } |
input.provider |
router::provider::on-changed |
the registry changes: a declaration, a discovery result or an availability flip | { "provider": "<id>", "op": "register" | "discovery" | "available" | "unavailable" } |
input.provider, input.op |
router::on-ready |
the router is ready: each observer receives the event once the router has booted and the observer is accepted, once per router start and once more when its worker reconnects within the observer grace period | {} |
— |
Delivery is iii's: every call reaches each observer whose filters match, and
the router calls a function only while it is observed. The ready event is a
call for one observer: the router calls router::on-ready with metadata
{ "for_observer": "<observer id>" }, and only that observer matches it, so
its target receives the Observed envelope like any other event. Observe
with capture: ['output'] so the envelope carries the event at output:
iii.registerFunction('my-worker::on-models-changed', async ({ output: event }) => {
console.log('catalog changed:', event); // { provider, count }
return {};
});
iii.observe(
{
function_id: 'router::models::on-changed',
filters: { 'input.provider': { $eq: 'anthropic' } },
capture: ['output'],
},
'my-worker::on-models-changed',
);The router itself observes configuration::on-updated (its entry,
router::on_config_changed), engine::on-functions-changed
(router::on_functions_changed) and, when the optional secrets worker is
installed, secrets::on-changed (router::on_secret_changed). The last
observer waits pending until the secrets worker registers its function.
Writing a provider worker
A provider worker must:
- Register
provider::<id>::streamhonouring the channel-writer contract: forward upstream output asAssistantMessageEventframes into thewriter_refit receives, ending with one terminal frame. - Declare itself at startup via
router::provider::register— retrying with backoff until acknowledged (covers provider-before-router boot order) — and re-declare on everyrouter::on-readyevent. Observe it once at boot (Rust providers useprovider_scaffold::router_client::observe_router_readywith theirprovider::<id>::on_router_readyhandler as the target); the observer outlives router restarts. The declaration may carryicon_svg: one self-contained, monochrome<svg>document (squareviewBox, no fixed size, at most 32 KiB). The router echoes it inrouter::provider::listand the console paints it as acurrentColormask beside the provider's models; a missing or malformed mark falls back to the provider's initial. The declaration may also carrydefault_models, the provider's recommended starting models in preference order (the current mid-range model first, then its predecessors), anddefault_thinking_level, the level to pair with it; see Default model. - Resolve credentials per request via
router::provider::resolve; never read keys directly. A response withcredential_source: "secret"and no credential is an unresolvablesecret://orenv://reference: report itscredential_error, do not substitute a key of your own. - Treat closure of its stream channel as cancellation: abort the upstream request and stop writing frames.
- Map upstream failures to the shared
ErrorKindtaxonomy on itserrorframes. Transport retries (429 / 5xx / connect) are the router's job, not the provider's.
A provider may also register provider::<id>::count_tokens ({model, system_prompt?, tools?, messages, thinking_level?, response_format?} →
{model, tokens, estimator}; the last two are the turn's router::chat
values, sent only when set) to serve
router::count_tokens: an exact count from a provider metering API
(estimator: "provider") or a local tokenizer estimate (estimator: "tiktoken"). Providers without it simply make router::count_tokens return
a typed router/no_token_counter error for that provider.
Default model
router::provider::list reports one default_model per provider so that a
caller with no model choice of its own (a fresh harness::send naming only
a provider, a console opening its first chat) starts on something sensible.
It is resolved at read time against the provider's current catalog slice:
- the model variant first. A variant is the first all-letter id segment
after the family (
terraingpt-6.1-terra,sonnetinclaude-sonnet-5-5), ranked in the orderdefault_modelsfirst names it. Within a ranked variant the newest version in the slice wins, listed or not, so a newer Luna never beats a Terra when Terra is listed first. A model with a ranked variant counts even outside the first preference's family, as long as it has the samevendor/prefix, so Copilot ranks GPT and Claude models in one list; - ids with no ranked variant come last: the first id in the declared
default_modelsthe slice holds, else the slice model sharing the longest id prefix with the first preference, as long as they share the family (the id up to its first-:claude,gpt,codex/gpt). On a tie the highest version wins (gpt-4.1overgpt-4o), then the plain id over a dated snapshot or a-minivariant; - otherwise absent, and the caller keeps its previous behaviour.
A provider that declares no default_models (local model servers, speech
providers) never reports one. default_thinking_level is copied from the
declaration; absent means the caller should omit the level and let the
provider apply its own default.
A declaration may also carry context_overflow_hint: one or two plain
sentences a console shows under a context-overflow failure on one of the
provider's models, for causes compacting cannot fix (a self-hosted server
started with a small window, say), optionally with a link to the provider's
docs. router::provider::list reports it verbatim; absent means the console
shows only its generic steps.
A provider that works without a key (a local server started without one)
declares credential_optional: true. The router then reports it
configured when no credential resolves; a key that is set is still
resolved and sent. Absent or false means a key is required.
Thinking levels
router::chat takes thinking_level: off, minimal, low, medium,
high or xhigh. Each provider maps a level onto its own knob and warns
when a model cannot honour it. off asks for no reasoning at all and is
honoured only where the model's descriptor says supports_thinking_off: true (a native switch exists: OpenAI reasoning_effort: none on the
models that take it, DeepSeek and GLM thinking.type: disabled, Anthropic
Sonnet 5.5 between_tools). false means the model always reasons or the
provider has no switch; absent means the provider does not know. Consoles
offer, hide or disable the Off choice accordingly. A provider that cannot
switch reasoning off sends its lowest effort instead and reports a warning.
Speech providers
A speech provider is the same kind of worker with no chat stream: it
declares its models with a speech block and serves one or both speech
surfaces. router::models::list hides speech models from chat pickers
unless asked (modality: "stt" | "tts" | "any"), and router::chat
refuses them with a message that names the right door.
{ "id": "scribe-v1", "provider": "elevenlabs", "context_window": 0, "max_output_tokens": 0,
"speech": { "modality": "stt", "languages": ["en", "hi"], "streaming": true } }| Function | Request | Response |
|---|---|---|
provider::<id>::transcribe |
{model?, audio_base64, mime, language?, prompt?} |
{model, text, segments[{text, start_secs?, end_secs?}]?, language?, duration_secs?} |
provider::<id>::speak |
{model?, text, voice?, format, language?, speed?} |
{model, audio_base64, mime, voice?, duration_secs?} |
model arrives as the id the caller named, or null for the provider's
default. format is what the caller asked for (mp3, wav, pcm16,
opus); answer with the container you produced and say so in mime.
Credentials resolve through router::provider::resolve exactly as for
chat providers.
The first real provider implementing this protocol is
provider-anthropic/ — useful as a reference
implementation alongside the scripted provider in the integration tests.
provider-openai/ follows the same structure for the
OpenAI Chat Completions API (native structured output, reasoning_effort).
provider-opencode-go/ follows the same structure for the
OpenCode Go Chat Completions API (curated model metadata, reasoning_effort).
Local development & testing
cargo test # unit suite, no engine needed
cargo test --test integration # engine-backed suite; self-skips without an engineThe integration suite spawns a throwaway engine per test when iii is on
PATH (or III_ENGINE_BIN points at a binary) and covers the chat relay,
cancellation, abort, restart recovery, registration token gating, paste-a-key
discovery (literal keys, secret:// and env:// references, against a fake
secrets worker), and event delivery end to end.
To run the worker locally against an engine:
cargo run -- --url ws://127.0.0.1:49134--url defaults to ws://127.0.0.1:49134 and honours the III_WS_URL
environment variable when the flag is not set. --config is accepted per
the standard worker CLI but ignored with a warning — operator config lives
in the engine's llm-router configuration entry (see Configuration above).