Daleen

Know the system. Confirm the change.

Daleen briefs your AI coding agents on your real database, your own standards, and what a change will touch before they write code, and keeps the record of what each task touched and cost.

Claude Code, Codex, Copilot and Cursor call Daleen first, through MCP. The same brief reaches every agent: the schema with ownership and grants, your standards compiled into pass or fail checks, the callers and tables a change will touch, the tests that must run, and the cost of the task.

Start with the free Diagnostic See the nine functions

The agent is fast. It does not know your database, it ignores standards nobody serves it, and nobody can say what the task cost.

Every coding agent ships with memory, rules files and a spend dashboard. None of them knows which grant your migration will widen, which of your standards outranks the other, or which tests cover the table it is about to change. Daleen answers those questions from the server, before the code is written, in the same words for every agent.

Nine functions

Each is one MCP tool family the agent calls. Each answers in well under a second.

  1. Startup Pack

    The brief every agent calls first: repository, commit, database access, standards and their authority, open items, and after a loss, where it left off.

  2. Compaction Protection

    When an agent's context is compacted, Daleen hands it a Resume Brief so work is not redone or forgotten.

  3. Schema Intel

    The real database, served to the agent: tables, columns, constraints, who owns each table, who can reach it, and the grants that drifted since last time.

  4. Standards Check

    Your own standards documents, indexed with authority and contradictions, searched by topic, and compiled into pass or fail checks the agent must satisfy before coding.

  5. Impact Analysis

    What a change touches before it is made: callers, possible callers, imports, tables and cascades, routes and tests.

  6. Regression Testing

    A registry of re-runnable tests that grows every time an agent touches something uncovered, and the list of what must pass before handoff.

  7. Work Record

    What the agent did, decided and left undone, filed as it goes, and whether the session is ready to hand off.

  8. Value Evidence

    The proof Daleen keeps about itself: what it predicted a change would touch, what the commit actually touched, and the fallout, with and without Daleen.

  9. Cost Ledger

    What each task cost in AI spend, per provider, per Work Packet, per session, per account. Exact for Codex and API-key agents; by reconciliation for the rest.

How it works

  1. Context before coding

    The agent opens with the Startup Pack. It gets the repository at this commit, the real schema with ownership and grants from Schema Intel, and the standards that apply from Standards Check, ranked by authority, with contradictions named.

  2. Impact before change

    Before it edits, the agent asks Impact Analysis what the change will touch and Regression Testing which tests must pass. The migration that widens a grant is seen before it is written, not in review.

  3. Validation before handoff

    The agent runs the required tests, files the Work Record, and the Cost Ledger prices the task. Value Evidence compares what was predicted with what the commit actually touched. If the context is compacted along the way, Compaction Protection brings the agent back.

The free Diagnostic

One repository and a read-only database connection. It takes about a day and produces one report: where grants have drifted, where agents code against standards nobody serves them, and what your agent sessions cost. If it is worth something, the service follows. If not, keep the report.

Ask for a Diagnostic

Security and retention

Never stored

  • Prompt bodies
  • Request and response bodies
  • File contents
  • Card numbers and CVV
  • Bank data
  • Passwords
  • Connection strings
  • API keys

Code maps hold paths and symbol names, never file bodies. The gateway keeps hashes and character counts. A publish that carries a connection string whole is refused.

Retention

Agent events keep 90 days of detail and day-grain token counts after that. Code maps keep the newest three per workspace. The tool call log keeps argument names and sizes, not values, for 180 days. Your own standards documents are the one class of customer text Daleen holds, replaced on every publish. A deletion request removes every row scoped to your workspaces, the rows naming your people, and your standards content within 30 days; nothing is restored from backup without the same deletion. Retention standard dated 2026-11-03.

Measured latency

Every tool has a budget in a published Performance Standard, measured over the network against the live service on 2026-11-02:

Full call time including network, cold then warm, release 05e2ccc
ToolColdWarm
Startup Pack1.26 s0.60 s
Schema Intel (table intel)0.29 s0.22 s
Standards Check (search)0.29 s0.21 s
Impact Analysis0.63 s0.52 s
Regression Testing (required tests)0.30 s0.21 s
Work Record (record work)0.44 s0.24 s

What we claim, and what we do not

A study is in progress. Until it reaches its sample we publish no productivity number: no "times faster," no percentage, no result extrapolated from one codebase to yours. What we can show today is the Diagnostic on your own repository and the measured latency above.

Price

$1,000 to $2,000 a month per company, on a ninety-day term.

The floor is $1,000 a month. Where a company lands in the range depends on the repositories and databases connected. Priced per company, not per seat: every engineer and every agent on the account is included. The Diagnostic is free and comes first.

Contact

Write to Jim Daleen. One email, no form.

[email protected] (alias being set up; until it is live, [email protected] reaches the same desk)