> ## Documentation Index > Fetch the complete documentation index at: https://smithers.sh/llms.txt > Use this file to discover all available pages before exploring further. # Platform Capabilities > Everything else you can ask for: plug in your own logging, get alerted, rewind runs, teach memory across runs, tune prompts, manage accounts and cost, share packs. Smithers is mostly built for your agent, not for you: your agent reads the [Technical API](/introduction) and does the operating. But knowing what Smithers *can* do helps you put its most powerful features to work, and that is what this page is for. [What You Can Do](/guide/what-you-can-do) catalogs the workflows Smithers runs. This page catalogs everything *around* those runs: the platform capabilities a run picks up for free. You don't operate any of this by hand; you ask your agent, exactly as in [Talk to Your Agent](/guide/talk-to-your-agent), and it runs the command or writes the config shown. Each row pairs what you say with what Smithers does under the hood. The deeper links lead into the Technical API; point your agent at them instead of reading them yourself. The few genuinely human steps (connecting your harness, signing up for semantic memory) live in the Set up pages, starting at [Set Up with Claude Code](/guide/setup/claude-code). ## Plug in your own logging Smithers logging is structured and replaceable, whether you run the CLI or embed Smithers in your own server. | You say | Smithers runs under the hood | | --------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------- | | "Ship Smithers logs into our logging stack as JSON, warnings and up." | `SMITHERS_LOG_FORMAT=json` + `SMITHERS_LOG_LEVEL=warn`; formats are `logfmt`, `json`, `pretty`, `string`. | | "We embed Smithers in our server and already have a logger. Use ours, not yours." | `createSmithersObservabilityLayer({ installLogger: false })` keeps your own logger installed ([Observability API](/reference/observability)). | | "Mirror every run event and failure into our own sink." | The `onProgress` and `onError` run callbacks, which receive every event and failure report as plain objects. | | "Send traces and metrics to our Grafana / Datadog." | OTLP export via `SMITHERS_OTEL_ENABLED=1` + `OTEL_EXPORTER_OTLP_ENDPOINT`; Prometheus text at the server's `GET /metrics`. | | "Give me a local dashboard with traces so I can poke around." | `bunx smithers-orchestrator observability`: a Grafana + Prometheus + Tempo + OTLP collector stack via Docker Compose. | ## Get alerted | You say | Smithers runs under the hood | | ----------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | "When the nightly release run fails, page the on-call with the runbook link." | An `alertPolicy` on the workflow (ownership, severity, runbook, reactions), with durable alert rows via `bunx smithers-orchestrator alerts` ([Alerting](/guides/alerting)). | | "Send me a daily digest of my runs on Telegram." | The [Telegram integration](/integrations/telegram) and the `telegram-daily-digest` pack workflow. | ## Go back in time Every step of every run is checkpointed, so history is something you can act on, not just read. | You say | Smithers runs under the hood | | ------------------------------------------------------------------------ | --------------------------------------------------------------------------------------- | | "That refactor went sideways at step 4. Rewind to just before it." | `bunx smithers-orchestrator rewind` back to the good frame. | | "Fork the run from before the review step and try a different approach." | `bunx smithers-orchestrator fork`, then `replay` drives the branch forward. | | "Show me every checkpoint this run took and what each one changed." | `bunx smithers-orchestrator timeline`, `snapshots`, and `diff`. | | "Put my files back the way they were before that task ran." | `bunx smithers-orchestrator restore` / `revert` reset the working tree to a checkpoint. | See [time travel in Recipes](/recipes#time-travel-fork-replay-diff) for the full loop. ## Take over mid-run | You say | Smithers runs under the hood | | ---------------------------------------------------------------------- | -------------------------------------------------------------------------------------------- | | "Hand me that agent session; I'll drive it from my terminal." | `bunx smithers-orchestrator hijack` hands the run's live agent session to you, resumable. | | "Pause the migration run: finish what's in flight, start nothing new." | `bunx smithers-orchestrator pause` parks it durably; resume later with `up --resume RUN_ID`. | | "The run is waiting on the deploy; tell it the deploy finished." | `bunx smithers-orchestrator signal` delivers the event a waiting run is listening for. | | "Did any run ask me a question? Show me, and let me answer." | `bunx smithers-orchestrator human` lists and resolves durable questions raised mid-run. | | "If my overnight runs stall, resume them automatically." | `bunx smithers-orchestrator supervise` watches named runs and auto-resumes them. | ## Teach it and tune it | You say | Smithers runs under the hood | | --------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | "What has Smithers learned about this repo across runs?" | `bunx smithers-orchestrator memory`: cross-run memory facts, queryable ([Memory](/concepts/memory)). | | "Give it memory that recalls by meaning, not keywords." | A Hindsight endpoint via `HINDSIGHT_URL`. Signing up or self-hosting is the one step your agent can't do for you: [Set Up Semantic Memory](/guide/setup/semantic-memory). | | "Protect this workflow with regression cases we can run in CI." | `bunx smithers-orchestrator eval` over a JSON/JSONL suite, emitting a check-in-able report ([Eval Suites Quickstart](/guides/evals-quickstart)). | | "How did the last few runs score?" | `bunx smithers-orchestrator scores` ([Scorers](/reference/scorers)). | | "Make the reviewer prompt better, and prove it's better." | `bunx smithers-orchestrator optimize`: GEPA prompt optimization verified against your eval suite ([Workflow optimization](/guides/workflow-optimization)). | ## Accounts, quota, and cost | You say | Smithers runs under the hood | | -------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------- | | "Use my Claude subscription for the smart steps and my OpenAI key for the rest." | `bunx smithers-orchestrator agents`: register subscriptions and API keys, then route each task to the right one. | | "How much of my quota have runs burned this week?" | `bunx smithers-orchestrator usage` per registered account. | ## Share and extend | You say | Smithers runs under the hood | | ---------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------- | | "Install that team's review pack." | `bunx smithers-orchestrator add` from GitHub, npm, or a local file. | | "Publish our pack so other teams can install it." | `bunx smithers-orchestrator share` opens the pull request for you. | | "Update the packs we've installed." | `bunx smithers-orchestrator packs update`. | | "Copy the built-in implement workflow into the repo so we can customize it." | `bunx smithers-orchestrator eject` copies the workflow plus its UI, prompts, and libraries. | | "Upgrade Smithers itself." | `bunx smithers-orchestrator update`, or `upgrade` for the agent-assisted path. | ## Quick jobs, no workflow | You say | Smithers runs under the hood | | -------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------ | | "Fix this one bug in the background and show me a live dashboard with the diff." | `bunx smithers-orchestrator oneshot`: one agent, one goal, chat/diff/hijack controls ([Oneshot](/guides/oneshot)). | | "Review this PR and give me a walkthrough I can actually read." | `bunx smithers-orchestrator review`: code review plus a story-form HTML walkthrough. | | "What happened in that run?" / "Why is it stuck?" | `bunx smithers-orchestrator what` / `why`: a narrated recap, or the exact blocker. | | "Ask Smithers how I'd do X." | `bunx smithers-orchestrator ask` answers from the installed docs and MCP server. | ## Run it your way | You say | Smithers runs under the hood | | ------------------------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------- | | "Run the risky steps in an isolated cloud sandbox." | Sandbox providers: Microsandbox, Daytona, Vercel, AWS, GCP ([Sandbox providers](/components/sandbox-providers)). | | "Let a teammate watch and approve from their laptop." | `bunx smithers-orchestrator gateway` serves the HTTP control plane; `token` mints short-lived access ([Gateway](/integrations/gateway)). | | "Keep run history in our Postgres, not a local SQLite file." | `bunx smithers-orchestrator migrate` moves the run store to PGlite or Postgres ([Database](/reference/db)). | ## Read next **Reference:** the workflow catalog these capabilities wrap around. **How-to:** the observe loop: list, follow, inspect, approve, cancel. **Technical API:** every command above, with flags, written for your agent.