The Agents page is the canonical home for everything persona-shaped. Every voice session in your org points at one of these rows; each agent owns its own set of tools. For the conceptual primer, see Concepts → Agents.Documentation Index
Fetch the complete documentation index at: https://docs.speko.dev/llms.txt
Use this file to discover all available pages before exploring further.
Creating an agent
Click Create agent, fill in:- Name — letters/digits/spaces/hyphens/underscores, ≤ 64 chars. Must be unique within your organization.
- System prompt — the agent’s instructions. The LLM sees this verbatim every session.
- TTS voice id (optional) — provider-specific id. Leave blank to let Speko’s router pick the provider’s default for the language.
- Language (required) — BCP-47 tag (
en,en-US,es-MX, etc.). - Optimize for (optional) —
latency/quality/cost— biases the router. - Stack preferences (optional) — per-layer allowlists (STT / LLM / TTS). Empty layers = no constraint. Chips are now grouped by vendor with one chip per known model — selecting Any Deepgram is equivalent to entering
"deepgram"in the API, while selectingnova-3under Deepgram is"deepgram:nova-3". See Concepts → Agents → Stack preferences for the merge semantics with per-call overrides. - Advanced LLM options (optional) —
temperature,maxTokens,model.
agent_a1b2c3d4e5f60718) appears at the top with a copy button — that’s what callers pass to POST /v1/sessions and to the SDK’s VoiceConversation.create({ agentId }).
The detail page
Each agent has three tabs:- Persona — the form you used to create it. Edits apply to the next session started against this agent; in-flight sessions keep the snapshot they were minted with.
- Tools (N) — the list of webhook tools registered under this agent. Add, edit, and delete from here. Tools are scoped per agent — different agents can register the same tool name without colliding.
- Sessions (N) — every voice session started against this agent. Read-only — the list is just a filtered view of
/sessionsscoped to this agentId. Click any row to drill into the per-session pipeline config and usage.
The Default agent
Every organization starts with a Default agent, seeded automatically at signup. It exists so the playground works on day one. You can edit its system prompt, voice, and intent freely; you cannot rename it or delete it.
Reaching it from the SDK
Once you have an id, the browser SDK call is:VoiceConversation for the full options and the legacy server-minted-session form.