Threads persist across crashes, deploys, and cold starts so live agent state survives process turnover.
Talon
The control plane for cloud native agents. Operate autonomous agent fleets at scale, securely.
The cloud native agent harness. Designed for AI-native companies operating critical agents in production.
Define agent templates, MCPs, schedules, knowledge, and namespaces once. Talon syncs your manifests to production in realtime.
Bindings, schedules, session memory, and recovery primitives ship together instead of being rebuilt around every agent.
Keep tools, memory, and execution boundaries scoped to the right environment so agents can operate safely in production.
Push agent specs to production in seconds. Talon's spec-driven configuration makes it easy to operate autonomous agents.
Agent specs stay declarative
Use AgentTemplates to declare your agent’s identity, capabilities, and tools. Talon’s runtime ensures the agent’s behavior stays consistent with its spec, even as you iterate on your implementation.
apiVersion: talon.impalasys.com/v1 kind: AgentTemplate metadata: name: overwatch definition: variables: - name: teamName type: string systemPrompt: | You are Overwatch, a book-keeping agent for the {{teamName}} team. Your job is to help the teams stay on top of their work by keeping track of their tasks, deadlines, and meetings. You are proactive and will send reminders and updates to them. capabilities: schedules: - inspect - create - update mcpServerRefs: - slack - google-docs - google-search - meetings
MCP transports stay explicit
MCP servers are declared globally so they can be referenced by agent specs to enhance their functionality.
apiVersion: talon.impalasys.com/v1 kind: McpServer metadata: name: parallel-search spec: transport: http target: https://search.parallel.ai/mcp args: [] headers: Authorization: Bearer sk-1234567890abcdef disabled: false
Session stays alive
Your frontend can connect directly to the Talon gateway using a secure, namespace-scoped token to send messages and stream responses in real time.
const sessionId = await createSession( 'https://gateway.talon.example.com', await getUserTalonToken(), "acme", "overwatch", ); const sessionUrl = 'https://gateway.talon.example.com/v1/ns/acme/agents/overwatch/sessions/019dbb39-b9b5-7533-a7dd-d367247ef612/message'; await fetch(sessionUrl, { method: "POST", headers: { "Content-Type": "application/json", Authorization: `Bearer ${token}`, }, body: JSON.stringify({ ns: "acme", agent: "overwatch", sessionId, message: "Review renewals due in 14 days.", }), cache: "no-store", }, );
Schedules trigger work
Schedules do not just launch jobs. They wake a specific agent and session pair so long-running workflows keep continuity.
name: renewal-check ns: acme spec: kind: cron cron: "0 14 * * 1-5" timezone: America/Los_Angeles target: agent: overwatch sessionMode: fresh inputMessage: | Review team meetings and deadlines for the week. Summarize any upcoming renewals and send reminders to the team. enabled: true
Knowledge stays reusable
Knowledge books and shared memory let agents reuse context, inherit playbooks, and exchange durable facts instead of recomputing the same state from scratch.
await fetch( `${gatewayUrl}/v1/namespaces/${encodeURIComponent("acme")}/knowledge`, { method: "POST", headers: { "Content-Type": "application/json", Authorization: `Bearer ${token}`, }, body: JSON.stringify({ ns: "acme", knowledge: { apiVersion: "talon.impalasys.com/v1", kind: "Knowledge", metadata: { name: "shared-playbooks", namespace: "acme", }, spec: { path: "playbooks/renewals.md", content: "# Renewal Playbook\n\nEscalate enterprise accounts when usage drops 25% week over week.", }, }, }), cache: "no-store", }, );
A bird's-eye view on your fleet with Sightline. A dashboard to explore namespaces, agents, sessions, schedules, templates, and MCPs in your deployment.
