Add migration for llm_trajectories table

[odilitime]

Introduces 0062_add_llm_trajectories_table.sql so the table exists in databases that only apply SQL migrations; journal entry aligns with Drizzle tracking.

Made-with: Cursor

---

Note

Medium Risk
Introduces a new database table with indexes and foreign keys; main risk is migration/runtime impact on large databases and potential downstream reliance on the new schema.

Overview
Adds SQL migration 0062_add_llm_trajectories_table.sql to create a new llm_trajectories table for storing LLM request/response text, token/cost accounting, latency, success/error state, and JSON metadata.

The migration also adds several query indexes and foreign key constraints to organizations (cascade delete), users, and api_keys (set null), and updates Drizzle’s migrations journal (meta/_journal.json) to register migration 0062_add_llm_trajectories_table.

^{Reviewed by Cursor Bugbot for commit 0f203f7. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
eliza-cloud-v2	Ready	Preview, Comment	Apr 11, 2026 0:27am

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 9341e464-23b2-4a7f-be62-4e377e57c0ed

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch odi-db

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[Copilot]

Pull request overview

Adds a new Postgres migration so environments that apply raw SQL migrations (and rely on Drizzle’s migration journal) create the llm_trajectories table used for storing LLM call prompt/response trajectories and related metrics.

Changes:

• Add 0062_add_llm_trajectories_table.sql to create llm_trajectories, its indexes, and foreign keys.
• Update packages/db/migrations/meta/_journal.json to register migration 0062_add_llm_trajectories_table for Drizzle tracking.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
packages/db/migrations/meta/_journal.json	Registers the new 0062 migration in Drizzle’s journal so it is tracked/applied in order.
packages/db/migrations/0062_add_llm_trajectories_table.sql	Creates the llm_trajectories table with indexes and FK constraints to organizations, users, and api_keys.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[claude]

PR Review: Add migration for llm_trajectories table

The migration is well-structured overall — uses IF NOT EXISTS guards, DO blocks for idempotent FK constraints, and the SQL matches the Drizzle schema file. A few issues worth addressing:

---

Bugs / Data Integrity

1. _journal.json loses its trailing newline

The PR removes the final newline from _journal.json. This can cause noisy diffs and may break POSIX-compliant tooling. The trailing newline should be preserved.

2. Derived columns without CHECK constraints

total_tokens and total_cost are computed values (input + output) but have no constraints enforcing this invariant. Callers can silently write inconsistent totals. Suggested additions to the migration:

ALTER TABLE "llm_trajectories"
  ADD CONSTRAINT "chk_total_tokens" CHECK ("total_tokens" = "input_tokens" + "output_tokens");
ALTER TABLE "llm_trajectories"
  ADD CONSTRAINT "chk_total_cost" CHECK ("total_cost" = "input_cost" + "output_cost");

If totals are always computed server-side and never caller-supplied, consider dropping the derived columns in favour of generated columns or a view instead.

3. request_id has no index and no uniqueness guarantee

If request_id is a correlation/trace ID for lookups or deduplication, it needs a unique constraint or at minimum an index. A lookup by request_id is currently a full table scan:

CREATE UNIQUE INDEX IF NOT EXISTS "llm_trajectories_request_id_idx"
  ON "llm_trajectories" ("request_id")
  WHERE "request_id" IS NOT NULL;

---

Performance

4. No index on user_id

Indexes cover org+created, org+model, purpose, and org+purpose+created — but not user_id. Querying all LLM calls for a specific user (usage dashboards, billing) will require a full sequential scan:

CREATE INDEX IF NOT EXISTS "llm_trajectories_user_id_idx" ON "llm_trajectories" ("user_id");

5. Unbounded TEXT columns for prompt content

system_prompt, user_prompt, and response_text are unconstrained text. With modern large context windows, individual rows can be 500KB-2MB. Under production load the table will grow very fast. Consider partitioning by created_at (monthly) before the table gets large, and whether full prompt text is required or if a truncated version suffices for fine-tuning.

---

Security

6. Sensitive data stored in plaintext

This table will hold raw system prompts and user conversations for every LLM call — potentially including confidential business logic (system prompts) and user PII. Recommend: row-level security scoped to organization_id, a data retention/TTL policy to prevent unbounded growth, and evaluating whether system_prompt should be encrypted at rest.

---

Process

7. Schema file pre-dates the migration

Per CLAUDE.md, schema + migration should be committed together. packages/db/schemas/llm-trajectories.ts was already on main before this PR, creating a window where the schema definition existed but the migration had not been applied. Worth noting in the PR description why the migration was split, to avoid setting a precedent.

---

Summary

Severity	Issue
Medium	Derived columns (total_tokens, total_cost) lack CHECK constraints
Medium	request_id is unindexed — full scan for lookups
Medium	No index on user_id
Low	_journal.json loses trailing newline
Low	Unbounded TEXT prompt columns — partition/retention plan needed
Low	Raw prompt data in plaintext — access controls recommended
Nit	Schema-before-migration pattern deviates from CLAUDE.md workflow

The migration will apply cleanly and the SQL matches the Drizzle schema, so nothing will break. The derived-column consistency and missing indexes are the most important fixes before this table sees production traffic.