feat: run Data on a self-hosted Letta agent instead of a Zo persona

[EthanThatOneKid]

Closes the third open item in #9: Data stops being a Zo persona and becomes the first self-hosted Letta agent on the Zo host.

What changes

• Brain. channels/http/ turns a question into one headless turn of the local agent (letta --backend local --agent <id> -p <question> --output-format json) and reuses the conversation_id it returns for the next question on the same session. Continuity no longer needs a Cloud round trip.
• Prompt stays in one place. agent/instructions.md is still the source of truth; scripts/sync-persona.ts (npm run sync-persona, --dry-run supported) renders it into the agent's memory as system/persona.md and commits it. Covered by lib/sync-persona.test.ts.
• Provider key. The Letta harness reads only the canonical name, so the service exports DATA_OPENROUTER_API_KEY as OPENROUTER_API_KEY at startup.
• Serialized turns. The local backend is a file-backed store, so two turns must never interleave; they queue in order.
• Rollback. DATA_BRAIN=zo answers from the persona again. Both paths share framing and session bookkeeping, and /health now reports the brain, the agent id, and the resolved CLI path.

Verification

Run from a service-like environment — env -i with only HOME, PATH, the secret file, and DATA_LETTA_BIN:

• /health → brain: letta, lettaBin: /usr/bin/letta, agent id reported.
• A cited answer named channels/http/index.ts lines 356-357 (POST /ask) and readJsonBody lines 289-300; both checked against the file by hand.
• A second question on the same session continued the conversation (same conversation_id, contextual answer).
• The agent's memory repo holds system/persona.md seeded from agent/instructions.md (commit 190caa4), and a live turn answered "who are you" in Data's voice with its read-only boundary.
• npm run typecheck clean; npm test 17/17.

[wazoocomputer]

This pull request, "feat: run Data on a self-hosted Letta agent instead of a Zo persona" (#10), transitions the Data agent from a Zo persona to a self-hosted Letta agent. This change is driven by the goal to run Data on a local agent, improving continuity and streamlining prompt management. The core changes involve updating how Data's "brain" processes questions, centralizing the agent's prompt, and adapting to the new Letta-based execution environment.

Here's a breakdown of the changes:

File Path	Change Type	Summary
.gitignore	modified	Adds .letta/ to gitignore to exclude runtime state.
AGENTS.md	modified	Updates documentation to reflect the new self-hosted Letta agent architecture and prompt management.
README.md	modified	Updates the main README to describe Data's new identity as a self-hosted Letta agent and how it runs.
agent/README.md	modified	Clarifies how Data's prompt is managed within the self-hosted Letta agent.
channels/discord/README.md	modified	Updates description to align with the new brain architecture.
channels/http/README.md	modified	Details the new data-http service, its endpoints, and environment variables for the Letta agent.
channels/http/index.ts	modified	Implements the core logic for routing questions to the self-hosted Letta agent, including conversation management and environment variable handling.
lib/sync-persona.test.ts	added	Adds tests for the sync-persona script.
lib/sync-persona.ts	added	Implements the sync-persona script for seeding the agent's memory.
package.json	modified	Updates dependencies or scripts.
scripts/sync-persona.ts	added	Adds a script to synchronize the agent's persona.
services/README.md	modified	Minor update to services documentation.
services/http-api.md	modified	Updates documentation for the HTTP API to reflect the new architecture.

Reviewers should start by examining channels/http/index.ts to understand how questions are now routed to the self-hosted Letta agent and how conversations are managed.