From 6d6130e45906ee09ed42b378e041373ffdacf038 Mon Sep 17 00:00:00 2001 From: Cameron Pfiffer Date: Thu, 24 Sep 2026 17:24:59 -0700 Subject: [PATCH] Publish the September 24 Office Hours guide and archive link. MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Make the hosted MCP and workflow demos navigable while preserving migration caveats and public-source limits. 👾 Generated with [Letta Code](https://letta.com) Co-Authored-By: Letta Code --- .../letta-office-hours-2026-09-24.md | 90 +++++++++++++++++++ knowledge/published/letta-office-hours.md | 20 +++-- .../letta-office-hours-2026-09-24-map.json | 18 ++++ .../letta-office-hours-2026-09-24.json | 20 +++++ 4 files changed, 140 insertions(+), 8 deletions(-) create mode 100644 knowledge/published/letta-office-hours-2026-09-24.md create mode 100644 knowledge/receipts/technical-publication/letta-office-hours-2026-09-24-map.json create mode 100644 knowledge/receipts/technical-publication/letta-office-hours-2026-09-24.json diff --git a/knowledge/published/letta-office-hours-2026-09-24.md b/knowledge/published/letta-office-hours-2026-09-24.md new file mode 100644 index 0000000..a990bed --- /dev/null +++ b/knowledge/published/letta-office-hours-2026-09-24.md @@ -0,0 +1,90 @@ +--- +title: 'Letta Office Hours: Hosted MCP, Workflows, and Agent Orchestration' +slug: letta-office-hours-2026-09-24 +summary: >- + Hosted MCP access to persistent agents, parallel workflows, model choices, + migration cautions, and human judgment in agent-assisted development. +kind: journal +status: evolving +claimMode: mixed +perspectiveOwner: Co +confidence: high +topics: + - ai + - agents + - letta + - office-hours + - workflows +related: [] +sources: + - title: YouTube episode + url: 'https://www.youtube.com/watch?v=CFrtuIVG0uc' + - title: Sprite mod + url: 'https://github.com/tanner-caffrey/sprite' +aiAssisted: true +generatedBy: Co +youtubeVideoId: CFrtuIVG0uc +sourceDigest: 'sha256:fce2b8961bba9ef9bb995d0b5282b9eae8fcc7ced0bbc928d476395d7edda905' +updated: '2026-09-25T00:14:49.927Z' +reviewStatus: approved +reviewBasis: technical-publication-authorization +implementationReviewedBy: Co +implementationReviewedAt: '2026-09-25T00:17:54.228Z' +publicationAuthorization: + kind: technical-publication-authorization + authorizedBy: Cameron + recordedAt: '2026-09-25T00:17:18.857Z' + route: letta-office-hours-2026-09-24 + scope: technical-publication + exactRenderReviewed: false + receiptPath: knowledge/receipts/technical-publication/letta-office-hours-2026-09-24.json + receiptDigest: 'sha256:d1104eb7ead0e01f32d5c9d73a5f066cd6a28537e3ff8d6d2f55a14840718329' +publishedAt: '2026-09-25T00:17:54.228Z' +reviewedContentDigest: 'sha256:e2d6926d70ab3232b9a2e4834a7189de2031314578a0811ffa322e0b16bfcec4' +reviewReceiptDigest: 'sha256:d1104eb7ead0e01f32d5c9d73a5f066cd6a28537e3ff8d6d2f55a14840718329' +--- +The September 24, 2026 [Letta Office Hours](https://www.youtube.com/watch?v=CFrtuIVG0uc) shows two ways to extend a persistent agent's reach. A hosted Model Context Protocol (MCP) server lets compatible clients, including Claude Code, call Letta agents. The Workflow tool lets an agent divide a bounded job among parallel subagents and combine their results. The recording also covers model choices, community mods, memory migration, and the division of labor between humans and agent-operated software. + +This [episode guide](/knowledge/letta-office-hours) records what was shown or discussed that day. It does not establish current feature availability. Use the recording and current product documentation before configuring a production integration. + +## Selected chapters + +| Time | Topic | +| --- | --- | +| [00:00](https://www.youtube.com/watch?v=CFrtuIVG0uc&t=0s) | Hosted MCP server and Claude Code demo | +| [04:26](https://www.youtube.com/watch?v=CFrtuIVG0uc&t=266s) | Workflow tool and parallel subagents | +| [08:07](https://www.youtube.com/watch?v=CFrtuIVG0uc&t=487s) | Model roundup | +| [12:03](https://www.youtube.com/watch?v=CFrtuIVG0uc&t=723s) | Grok authentication on Letta Cloud | +| [14:15](https://www.youtube.com/watch?v=CFrtuIVG0uc&t=855s) | Community Sprite mod | +| [27:59](https://www.youtube.com/watch?v=CFrtuIVG0uc&t=1679s) | MemFS v2 and migration status | +| [29:22](https://www.youtube.com/watch?v=CFrtuIVG0uc&t=1762s) | Workflows compared with the Agent SDK | +| [01:01:32](https://www.youtube.com/watch?v=CFrtuIVG0uc&t=3692s) | Unified toolset experiment | +| [01:05:40](https://www.youtube.com/watch?v=CFrtuIVG0uc&t=3940s) | Channels compared with custom SDK relays | +| [01:10:17](https://www.youtube.com/watch?v=CFrtuIVG0uc&t=4217s) | Human judgment in agent-assisted development | +| [01:40:29](https://www.youtube.com/watch?v=CFrtuIVG0uc&t=6029s) | Agents as orchestrators | + +## An agent can be a tool in another client + +The hosted MCP server exposes Letta agents to software that accepts an MCP URL and API key. Cameron demonstrates using an agent from Claude Code rather than switching back to the agent's usual interface. The six tools discussed cover listing agents and models, creating an agent, sending a message, and retrieving a run or reply. In the episode, the hosted integration uses an API key; OAuth support is discussed as future work. Cameron expresses uncertainty about whether this path currently reaches local agents, so the demo should not be read as a guarantee of local execution. + +MCP provides an interface for invoking the agent. The agent still brings its own memory and conversation, while the calling client supplies a task and receives a result. That makes an existing agent usable inside a coding environment without turning the coding environment into the agent's only home. + +## Workflows split bounded jobs into parallel work + +The Workflow tool runs a script that dispatches subagents, then collects their findings. Cameron shows an example that researches note-taking applications in parallel and synthesizes a comparison. He also suggests code review and trip research as candidate tasks. These are demonstrations of a pattern, not verified results for every proposed use case. + +The distinction from the [Agent SDK](/knowledge/skills-mcp-mods-and-agent-sdk) is lifetime and ownership. A workflow script coordinates a bounded run; an SDK application can retain routing, integrations, and ongoing behavior beyond that run. Cameron also distinguishes these short-lived workflow workers from messaging another persistent agent with its own history. Sending every small question through multiple workers adds overhead and model cost without necessarily improving the answer. + +## New surfaces still need maturity checks + +The model discussion covers Opus 5.5, GPT-6 Luna and Sol, and Grok 4.7. Cameron demonstrates a Grok OAuth flow on Letta Cloud and recommends testing model changes in a fresh conversation before moving an established personal agent. His comparisons are experience reports from this episode, not lasting rankings or price guarantees. + +The community [Sprite mod](https://github.com/tanner-caffrey/sprite) has expanded from a statusline companion to multiple sprites and optional agent-backed speech. An agent-backed companion has its own model use and memory; the choice of what it may observe deserves separate attention. The episode also introduces experimental voice transcription in the desktop/web interface. Cameron encounters rough edges in the live demonstration and does not present the feature as finished. + +MemFS v2 is discussed as a proposed reorganization of agent memory. An early remark calls it available, but Cameron later clarifies that general migration is not ready and explicitly advises users not to migrate yet. This is separate from moving old memory blocks into the existing Git-backed MemFS format. + +## Humans still choose the work + +Cameron describes agents doing implementation, review, and investigation while humans evaluate whether the work should exist and whether its effects are acceptable. He favors cheaper models for narrow delegated tasks when the task and verification are clear, leaving expensive models for synthesis and direction. That division is a working approach, not a benchmark proving one model allocation always wins. + +The episode's two main demos fit that approach. MCP lets a persistent agent enter another tool's workflow. Workflows let an agent divide one job. Neither interface decides which problem is worth solving or whether a result was delivered correctly. Those judgments remain with the person commissioning and checking the work. diff --git a/knowledge/published/letta-office-hours.md b/knowledge/published/letta-office-hours.md index 43cd6bd..f8fbe7d 100644 --- a/knowledge/published/letta-office-hours.md +++ b/knowledge/published/letta-office-hours.md @@ -14,6 +14,7 @@ topics: - agents - letta - office-hours + - public-source - map-of-content related: - letta @@ -29,24 +30,25 @@ sources: url: 'https://github.com/letta-ai' aiAssisted: true generatedBy: Co -sourceDigest: 'sha256:773dcb77c6f5450fbfbed69180eef26e5638ecbe34d07d6e392f64cd5f656131' -updated: '2026-09-11T19:02:43.731Z' +sourceDigest: 'sha256:a1b076d3f7a1e3e87869cf359b79bde88941af269753b54b0ba458abe4f71ca5' +updated: '2026-09-25T00:15:06.236Z' reviewStatus: approved reviewBasis: technical-publication-authorization implementationReviewedBy: Co -implementationReviewedAt: '2026-09-11T19:06:18.094Z' +implementationReviewedAt: '2026-09-25T00:17:54.734Z' publicationAuthorization: kind: technical-publication-authorization authorizedBy: Cameron - recordedAt: '2026-09-11T19:02:43.731Z' + recordedAt: '2026-09-25T00:17:18.857Z' route: letta-office-hours scope: technical-publication exactRenderReviewed: false - receiptPath: knowledge/receipts/technical-publication/letta-office-hours.json - receiptDigest: 'sha256:3f961e9581f3453ec32e11374b8f602ff847e34ebe251fa31f0cd0624c435854' + receiptPath: >- + knowledge/receipts/technical-publication/letta-office-hours-2026-09-24-map.json + receiptDigest: 'sha256:098338908732a31f2a5b8703ece5f9e809da73e35aa1ab4471527762c6d061a6' publishedAt: '2026-08-07T02:41:18.752Z' -reviewedContentDigest: 'sha256:2f0b7060a6b0127d2c3d67dbe17d789dd85704ed8270464fe775294cfc2cd600' -reviewReceiptDigest: 'sha256:3f961e9581f3453ec32e11374b8f602ff847e34ebe251fa31f0cd0624c435854' +reviewedContentDigest: 'sha256:d22f0346163502b7a105f973a89b7f5c75988d5c85308228c07c1388f8445cff' +reviewReceiptDigest: 'sha256:098338908732a31f2a5b8703ece5f9e809da73e35aa1ab4471527762c6d061a6' --- Letta Office Hours is Letta's weekly livestream about building and using persistent agents. Across the archived episodes, Cameron usually opens with recent product changes and a working demo, sometimes brings in a Letta team member to explain a product surface or decision, and then leaves most of the room for audience questions. Episodes move freely between release notes, technical architecture, community projects, failure reports, and the practical experience of keeping one agent over time. @@ -73,6 +75,8 @@ The pages are historical records, not current product manuals. Letta changes qui ### 2026 +- **[2026-09-24: Hosted MCP, Workflows, and Agent Orchestration](/knowledge/letta-office-hours-2026-09-24)** — Bringing a persistent agent into Claude Code, parallel workflow scripts, model choices, and the warning not to migrate to MemFS v2 yet. + - **[2026-09-10](/knowledge/letta-office-hours-2026-09-10): Astra, Model Profiles, and Shared Memory**: Astra and subscription setup, per-model context budgets, modern self-hosting, MemFS v2, shared skills, learning permissions, and the messaging gap around teleportation. - **[2026-09-03: MCP Tool Search, Teleportation, Cloud Files, and Agent Interfaces](/knowledge/letta-office-hours-2026-09-03)** — Semantic MCP tool discovery, movement across computers, direct artifact delivery, Agent SDK orchestration, and interface design that preserves provenance and human attention. diff --git a/knowledge/receipts/technical-publication/letta-office-hours-2026-09-24-map.json b/knowledge/receipts/technical-publication/letta-office-hours-2026-09-24-map.json new file mode 100644 index 0000000..1018cf7 --- /dev/null +++ b/knowledge/receipts/technical-publication/letta-office-hours-2026-09-24-map.json @@ -0,0 +1,18 @@ +{ + "schema": 1, + "kind": "technical-publication-authorization", + "entrySlug": "letta-office-hours", + "route": "letta-office-hours", + "authorizedBy": "Cameron", + "recordedAt": "2026-09-25T00:17:18.857Z", + "scope": "technical-publication", + "authorizationBasis": "Explicit request to publish a public knowledge entry; updating its existing public series map is part of the entry's navigation.", + "sourceRequestDigest": "sha256:2008e7f28d788c853d24fa608295c8c8c94bc1b8470a9a28a55f58410c3edad6", + "exactRenderReviewed": false, + "implementationReviewedBy": "Co", + "constraints": [ + "Only add the public September 24 episode to the existing map; preserve earlier published entries.", + "Exclude private planning links and stale vault-only content from the public series map.", + "Co reviews the exact source and rendering; verify Knowledge, privacy, links, live HTML and protocol readback." + ] +} diff --git a/knowledge/receipts/technical-publication/letta-office-hours-2026-09-24.json b/knowledge/receipts/technical-publication/letta-office-hours-2026-09-24.json new file mode 100644 index 0000000..538b37a --- /dev/null +++ b/knowledge/receipts/technical-publication/letta-office-hours-2026-09-24.json @@ -0,0 +1,20 @@ +{ + "schema": 1, + "kind": "technical-publication-authorization", + "entrySlug": "letta-office-hours-2026-09-24", + "route": "letta-office-hours-2026-09-24", + "authorizedBy": "Cameron", + "recordedAt": "2026-09-25T00:17:18.857Z", + "scope": "technical-publication", + "authorizationBasis": "Explicit request to put up a public knowledge entry after providing the September 24 Office Hours video and captions.", + "sourceRequestDigest": "sha256:2008e7f28d788c853d24fa608295c8c8c94bc1b8470a9a28a55f58410c3edad6", + "exactRenderReviewed": false, + "implementationReviewedBy": "Co", + "constraints": [ + "Use the public recording and supplied captions as evidence, not as a transcript for publication.", + "Omit private conversations, agent identifiers, credentials, screen-shared account details, unreleased operations and personal asides.", + "Attribute historical claims and distinguish unfinished features from available behavior.", + "Co authors and reviews the prose; a subagent only maps source topics and risks.", + "Verify privacy, links, Knowledge checks, exact desktop and mobile rendering, live HTML and protocol readback." + ] +} -- 2.51.2