Coding Agent Memory

Long-term, crash-safe memory for autonomous coding agents.

Industry	Level	Difficulty	Recommended mode
Software engineering	4 of 5	Expert	flat

Coming soon

A complete, runnable Coding Agent Memory reference implementation is in progress. The architecture and core code below already work with SrvDB 0.3.

The problem

• Agents forget decisions between sessions.
• Long runs crash and lose context.
• Repos change constantly.

How SrvDB powers Coding Agent Memory

1. Decisions, attempts and outcomes are stored in SrvDB.
2. Agents recall relevant memory per task.
3. persist() checkpoints survive crashes.
4. Per-repo databases.

What changes

Today	With Coding Agent Memory on SrvDB
Agents repeat mistakes	Memory of outcomes
Lost context on crash	Crash-safe checkpoints
Global confusion	Per-repo memory

How it works

1. Log agent steps.
2. Embed summaries.
3. Recall per task.
4. Persist checkpoints.
5. Prune with delete.

SrvDB features it uses

• Crash-safe persist(): after it returns, data survives a crash or power loss and reopens exactly as committed.
• Several databases: each database is a directory, so different data types or tenants stay separate.
• upsert: inserts new items and replaces changed ones, so the knowledge base stays current.
• delete and compact(): remove items immediately from search, then reclaim disk space.

Use cases

• Autonomous refactors
• Long migrations
• Test fixing
• Dependency upgrades
• Code review agents
• Incident agents

Core code

embed() is your embedding model (for example sentence-transformers, or a function that turns your records into feature vectors).

import json
import srvdb

db = srvdb.Database("./coding-agent-memory-db", dimension=384)

def ingest(items):
    """items: list of (id, text, metadata dict)"""
    db.upsert(
        ids=[i for i, _, _ in items],
        embeddings=[embed(t) for _, t, _ in items],
        metadatas=[json.dumps(m) for _, _, m in items],
    )
    db.persist()                      # crash-safe checkpoint

def recall_memory(task, k=5):
    hits = db.search(embed(task), k)
    return [(json.loads(db.get(i)), round(score, 3)) for i, score in hits]

Limits and honest notes

• Review agent changes before merging.

Build roadmap

1. Prototype: load a sample of real data with flat mode and check that searches return what an expert expects.
2. Measure: if you move to an approximate mode, compare its results with flat on the same queries (recall).
3. Pipeline: add continuous upsert and decide when to call persist().
4. Harden: test restarts and crashes, add logging, review every output path with domain experts.
5. Deploy: ship the database directory with the application or service.

All Software engineering projects

Level	Project	What it does
1 · Beginner	Code Search	Find code by what it does.
2 · Intermediate	Codebase Q&A	Ask how the system works and get answers citing files.
3 · Advanced	Bug Dedup Engine	Links duplicate bug reports and similar stack traces.
4 · Expert	Coding Agent Memory (this page)	Long-term, crash-safe memory for autonomous coding agents.
5 · Frontier	Air-gapped Dev Copilot	A self-hosted developer assistant for isolated environments.

See also: Use cases · Live examples (coming soon)