Reference — Temporality

Reference

Environment variables, ports, UI pages and links to the in-depth documents.

#Environment variables (essentials)

Variable	Default	Description
`TEMPORALITY_MODEL_BASE_URL` / `_ID` / `_API_KEY`	—	The default provider (required)
`TEMPORALITY_MODEL_MAX_OUTPUT_TOKENS`	`16384` (code) / `1024` (compose)	Generation limit; leave head room for reasoning models. The compose stack defaults to 1024
`TEMPORALITY_MODEL_TIMEOUT`	`180s`	Base HTTP request timeout for the provider
`TEMPORALITY_MODEL_STREAM_IDLE_TIMEOUT`	`90s`	Silence in the stream between tokens
`TEMPORALITY_MODEL_STREAM_NO_PROGRESS_TIMEOUT`	`4m`	The stream is alive, but overall progress has stalled
`TEMPORALITY_MODEL_TEMPERATURE` / `_REASONING`	—	Overrides for request parameters
`TEMPORALITY_MODEL_ACTIVITY_TIMEOUT`	`20m`	Ceiling for one model activity
`TEMPORALITY_MODEL_FIRST_TOKEN_TIMEOUT`	`2m`	First-token wait (adaptive, +60s per MB of prompt)
`TEMPORALITY_TOOL_ACTIVITY_TIMEOUT`	`16m`	Ceiling for a `run_tool` activity
`KERNEL_TOOL_MAX_SECONDS`	`900`	Ceiling for a single sandbox command
`KERNEL_BG_TOOL_MAX_SECONDS`	`3600`	Ceiling for a detached background command — `run_command` with `background: true` gets its own budget of up to an hour
`KERNEL_EMBEDDINGS_MODEL`	`intfloat/multilingual-e5-small`	Model of the knowledge semantic layer (in-process, rembed)
`KERNEL_EMBEDDING_WORKERS`	—	Parallelism of background embedding computations
`KERNEL_PROVIDER_MODELS_REFRESH`	`24h`	Background refresh cycle of the provider model catalog (`0` — disable)
`KERNEL_SUPERVISOR_TRIAGE_INTERVAL`	`2m`	How often failed runs are triaged (values outside the valid range fall back to the default — there is no off switch)
`KERNEL_SANDBOX_ROOT`	`./.sandbox` (code) / `/srv/temporality/sandbox` (compose)	Root of the persistent workspaces. Do not use `/tmp` — cleanup on node reboot will destroy the workspaces
`KERNEL_SANDBOX_PIDS` / `KERNEL_SANDBOX_SCRATCH_SIZE`	—	Sandbox container limits: processes and scratch size
`KERNEL_DOCKER_BINARY`	`docker`	Path to the docker CLI
`KERNEL_SESSION_TTL`	`12h`	Sliding TTL of a browser session
`KERNEL_SESSION_ABSOLUTE_TTL`	`168h`	Absolute ceiling for a session
`KERNEL_COOKIE_SECURE`	`false`	Secure flag on the session cookie (enable it behind HTTPS)
`KERNEL_WEB_SEARCH_PER_MIN` / `KERNEL_BROWSER_PER_MIN`	`30` / `60`	Rate limits for search and the browser
`KERNEL_WEB_FETCH_ALLOW_PRIVATE`	`false`	Allow `web_fetch` into private networks (SSRF protection; enable deliberately)
`KERNEL_SECRET_CACHE_TTL_SECONDS`	—	How long a resolved secret value lives in the kernel's memory
`KERNEL_MCP_COMMAND` / `_ARGS` / `_ALLOW` / `_APPROVAL` / `_SERVER_NAME`	—	External MCP server: command, arguments, allowed tools, approval
`KERNEL_STORAGE_DEFAULT_QUOTA_BYTES` / `KERNEL_STORAGE_SWEEPER`	—	Sandbox storage quotas and sweeping
`TEMPORALITY_API_TOKEN`	auto-provisioned	Bearer token kernel → journal
`TEMPORALITY_OPERATIONS_WINDOW`	`7d`	Window for operational rollups
`KERNEL_UI_URL`	—	Base URL of the UI (emails/links)
`ARTIFACT_SUBMIT_PER_DAY`	`10`	Daily quota of artifact review submissions per run
`ARTIFACT_REVIEW_TIMEOUT`	`72h`	Artifact review deadline: one reviewer reminder, escalation to the unit head after a second timeout; `0` disables the sweep
`KERNEL_SECRET_MASTER_KEY`	—	Secret encryption key (`openssl rand -hex 32`)
`KERNEL_AUTH_TOKENS` / `JOURNAL_AUTH_TOKENS`	empty	Bearer tokens (`token:user:role:projects`; empty = auth off)
`KERNEL_RUN_QUOTAS`	empty	Daily run quotas: `project:limit`, `*` — default
`KERNEL_MODEL_PRICES`	empty	Token prices: `model:prompt:completion`
`KERNEL_INTAKE_MAX_PER_HOUR`	`10`	Per-sender limit for channel messages
`KERNEL_INTAKE_RETENTION_DAYS`	`30`	How long inbound bodies are kept
`KERNEL_SEARXNG_URL`	`http://searxng:8080`	Meta-search for `web_search` (empty = off)
`KERNEL_BROWSER_CDP_URL`	lightpanda	Headless browser for `browser` (empty = off)
`KERNEL_CHANNEL__MATRIX__`, `__TELEGRAM__`, `__MATTERMOST__`, `__SLACKBOT__`	—	Channel transports

The full list with semantics is in the root README.md and docs/runbook.md.

#Ports

Service	Port
UI (debugger)	`3000`
Journal API	`8080`
Agent Kernel API	`8090`
Temporal UI	`8233`
SearXNG	`127.0.0.1:8889`

#UI pages

Path	Section
`/dashboard`	Dashboard: health, models, projects, quality, sagas, sandboxes
`/supervisor`	Supervisor queue: triage of failed runs, cases, skill candidates
`/workspace`	Agent chat
`/agents`	Runs and traces
`/knowledge`	Knowledge: the map, search, dossiers, contradictions
`/experience`	The experience timeline
`/storage`	Sandbox storage (files)
`/agent-config` · `/teams`	Agents and teams
`/skills` · `/artifacts`	Skills and artifacts
`/inbound` · `/triggers` · `/channels`	Inbound, triggers, channels
`/providers` · `/secrets` · `/sandbox`	Providers, secrets, sandboxes
`/users` · `/org`	Users and the org structure
`/mcp`	External MCP server status

#Glossary

Run — one agent execution over a task (a durable workflow).
Addendum — a user message appended to a running run.
Amendment — a channel message continuing an existing task.
Saga — tracked long-running work across several runs and agents.
Knowledge — a statement with provenance, a status and a history.
User fact — a durable statement about a person (preferences, role, ownership); context for agents, never a permission or an instruction.
Policy (Org Policy) — a set of restrictions on an org unit (models, MCP, budgets, network, sandbox, egress, intake, filesystem areas); inherited and merged restrictively.
Execution identity — the service identity of automated runs (webhooks, schedules, triggers) with its own allowed lists.
Hint — a compact excerpt of experience fed to the agent.
Skill — a versioned procedure with a contract and evaluations.
Artifact — a document in the registry with versions and human approval, loaded on demand.
Reconcile — manually resolving an uncertain operation (happened / did not happen).
Judge — the LLM model evaluating answers in the quality loop.

#Engineering documents

The architecture and internals are described in the repository's docs/ (on the site — the Engineering documents section): architecture.md (overview), event-model.md, knowledge-model.md, agent-quality-loop.md, living-skills.md, agent-delegation.md, inbound-messages.md, org-structure.md (org structure, policies, execution identity), sandbox-security-matrix.md, failure-model.md, failure-reconciliation.md, resilience-and-scale.md (fault tolerance and scale bounds: replay, workflow history, Continue-As-New), runbook.md (operations) and others.