alvis/adolf
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Split CLAUDE.md per official Claude Code recommendations

Browse Source 
CLAUDE.md: lean — commands, key conventions, fast tool guide, @ARCHITECTURE.md import
routecheck/CLAUDE.md: purpose, access paths, env vars, gotchas
openmemory/CLAUDE.md: tools, two Ollama instances, prompts, notes

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Alvis
2026-03-13 07:15:51 +00:00
parent eba805f787
commit 3ed47b45da
3 changed files with 74 additions and 160 deletions
Download Patch File Download Diff File Expand all files Collapse all files

183
CLAUDE.md
View File

@@ -4,175 +4,38 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
## Commands
**Start all services:**
```bash
# Start all services
docker compose up --build
```
**Interactive CLI (Docker container, requires gateway running):**
```bash
# Interactive CLI (requires services running)
docker compose --profile tools run --rm -it cli
# or with options:
docker compose --profile tools run --rm -it cli python3 cli.py --url http://deepagents:8000 --session cli-alvis
# Integration tests (run from tests/integration/, requires all services)
python3 test_health.py
python3 test_memory.py [--name-only|--bench-only|--dedup-only]
python3 test_routing.py [--easy-only|--medium-only|--hard-only]
# Use case tests — read the .md file and follow its steps as Claude Code
# e.g.: read tests/use_cases/weather_now.md and execute it
```
**Run integration tests** (from `tests/integration/`, require all Docker services running):
```bash
python3 test_health.py # service health: deepagents, bifrost, Ollama, Qdrant, SearXNG
## Key Conventions
python3 test_memory.py # name store/recall + memory benchmark + dedup
python3 test_memory.py --name-only # only name store/recall pipeline
python3 test_memory.py --bench-only # only 5-fact store + 10-question recall
python3 test_memory.py --dedup-only # only deduplication test
- **Models via Bifrost only** — all LLM calls use `base_url=BIFROST_URL` with `ollama/<model>` prefix. Never call Ollama directly for inference.
- **One inference at a time** — `_reply_semaphore` serializes GPU use. Do not bypass it.
- **No tools in medium agent** — `_DirectModel` is a plain `ainvoke()` call. Context is injected via system prompt. `qwen3:4b` is unreliable with tool schemas.
- **Fast tools are pre-flight** — `FastToolRunner` runs before routing and before any LLM call. Results are injected as context, not returned to the user directly.
- **Memory outside agent loop** — `add_memory`/`search_memory` are called directly, never passed to agent tool lists.
- **Complex tier is opt-in** — `/think ` prefix only. LLM classification of "complex" is always downgraded to medium.
- **`.env` is required** — `TELEGRAM_BOT_TOKEN`, `ROUTECHECK_TOKEN`, `YANDEX_ROUTING_KEY`. Never commit it.
python3 test_routing.py # all routing benchmarks (easy + medium + hard)
python3 test_routing.py --easy-only # light-tier routing benchmark
python3 test_routing.py --medium-only # medium-tier routing benchmark
python3 test_routing.py --hard-only # complex-tier + VRAM flush benchmark
```
## Adding a Fast Tool
Shared config and helpers are in `tests/integration/common.py`.
**Use case tests** (`tests/use_cases/`) — markdown skill files executed by Claude Code, which acts as mock user and quality evaluator. Run by reading the `.md` file and following its steps with tools (Bash, WebFetch, etc.).
1. Subclass `FastTool` in `fast_tools.py` — implement `name`, `matches(message) → bool`, `run(message) → str`
2. Add instance to `_fast_tool_runner` list in `agent.py`
3. The router will automatically force medium tier when `matches()` returns true
## Architecture
Adolf is a multi-channel personal assistant. All LLM inference is routed through **Bifrost**, an open-source Go-based LLM gateway that adds retry logic, failover, and observability in front of Ollama.
### Request flow
```
Channel adapter → POST /message {text, session_id, channel, user_id}
→ 202 Accepted (immediate)
→ background: run_agent_task()
→ asyncio.gather(
_fetch_urls_from_message() ← Crawl4AI, concurrent
_retrieve_memories() ← openmemory search, concurrent
_fast_tool_runner.run_matching() ← FastTools (weather, commute), concurrent
)
→ router.route() → tier decision (light/medium/complex)
fast tool match → force medium
if URL content fetched → upgrade light→medium
→ invoke agent for tier via Bifrost (url_context + memories in system prompt)
deepagents:8000 → bifrost:8080/v1 → ollama:11436
→ _push_stream_chunk() per token (medium streaming) / full reply (light, complex)
→ _stream_queues[session_id] asyncio.Queue
→ _end_stream() sends [DONE] sentinel
→ channels.deliver(session_id, channel, reply)
→ channel-specific callback (Telegram POST)
→ _store_memory() background task (openmemory)
CLI streaming → GET /stream/{session_id} (SSE, per-token for medium, single-chunk for others)
```
### Bifrost integration
Bifrost (`bifrost-config.json`) is configured with the `ollama` provider pointing to the GPU Ollama instance on host port 11436. It exposes an OpenAI-compatible API at `http://bifrost:8080/v1`.
`agent.py` uses `langchain_openai.ChatOpenAI` with `base_url=BIFROST_URL`. Model names use the `provider/model` format that Bifrost expects: `ollama/qwen3:4b`, `ollama/qwen3:8b`, `ollama/qwen2.5:1.5b`. Bifrost strips the `ollama/` prefix before forwarding to Ollama.
`VRAMManager` bypasses Bifrost and talks directly to Ollama via `OLLAMA_BASE_URL` (host:11436) for flush/poll/prewarm operations — Bifrost cannot manage GPU VRAM.
### Three-tier routing (`router.py`, `agent.py`)
| Tier | Model (env var) | Trigger |
|------|-----------------|---------|
| light | `qwen2.5:1.5b` (`DEEPAGENTS_ROUTER_MODEL`) | Regex pre-match or LLM classifies "light" — answered by router model directly, no agent invoked |
| medium | `qwen3:4b` (`DEEPAGENTS_MODEL`) | Default for tool-requiring queries |
| complex | `qwen3:8b` (`DEEPAGENTS_COMPLEX_MODEL`) | `/think ` prefix only |
The router does regex pre-classification first, then LLM classification. Complex tier is blocked unless the message starts with `/think ` — any LLM classification of "complex" is downgraded to medium.
A global `asyncio.Semaphore(1)` (`_reply_semaphore`) serializes all LLM inference — one request at a time.
### Thinking mode and streaming
qwen3 models produce chain-of-thought `<think>...</think>` tokens. Handling differs by tier:
- **Medium** (`qwen3:4b`): streams via `astream()`. A state machine (`in_think` flag) filters `<think>` blocks in real time — only non-think tokens are pushed to `_stream_queues` and displayed to the user.
- **Complex** (`qwen3:8b`): `create_deep_agent` returns a complete reply; `_strip_think()` filters think blocks before the reply is pushed as a single chunk.
- **Router/light** (`qwen2.5:1.5b`): no thinking support; `_strip_think()` used defensively.
`_strip_think()` in `agent.py` and `router.py` strips any `<think>` blocks from non-streaming output.
### VRAM management (`vram_manager.py`)
Hardware: GTX 1070 (8 GB). Before running the 8b model, medium models are flushed via Ollama `keep_alive=0`, then `/api/ps` is polled (15s timeout) to confirm eviction. On timeout, falls back to medium tier. After complex reply, 8b is flushed and medium models are pre-warmed as a background task.
### Channel adapters (`channels.py`)
- **Telegram**: Grammy Node.js bot (`grammy/bot.mjs`) long-polls Telegram → `POST /message`; replies delivered via `POST grammy:3001/send`
- **CLI**: `cli.py` (Docker container, `profiles: [tools]`) posts to `/message`, then streams from `GET /stream/{session_id}` SSE with Rich `Live` display and final Markdown render.
Session IDs: `tg-<chat_id>` for Telegram, `cli-<username>` for CLI. Conversation history: 5-turn buffer per session.
### Services (`docker-compose.yml`)
| Service | Port | Role |
|---------|------|------|
| `bifrost` | 8080 | LLM gateway — retries, failover, observability; config from `bifrost-config.json` |
| `deepagents` | 8000 | FastAPI gateway + agent core |
| `openmemory` | 8765 | FastMCP server + mem0 memory tools (Qdrant-backed) |
| `grammy` | 3001 | grammY Telegram bot + `/send` HTTP endpoint |
| `crawl4ai` | 11235 | JS-rendered page fetching |
| `routecheck` | 8090 | Local routing web service — image captcha UI + Yandex Routing API backend |
| `cli` | — | Interactive CLI container (`profiles: [tools]`), Rich streaming display |
External (from `openai/` stack, host ports):
- Ollama GPU: `11436` — all reply inference (via Bifrost) + VRAM management (direct)
- Ollama CPU: `11435` — nomic-embed-text embeddings for openmemory
- Qdrant: `6333` — vector store for memories
- SearXNG: `11437` — web search
### Bifrost config (`bifrost-config.json`)
The file is mounted into the bifrost container at `/app/data/config.json`. It declares one Ollama provider key pointing to `host.docker.internal:11436` with 2 retries and 300s timeout. To add fallback providers or adjust weights, edit this file and restart the bifrost container.
### Crawl4AI integration
Crawl4AI is embedded at all levels of the pipeline:
- **Pre-routing (all tiers)**: `_fetch_urls_from_message()` detects URLs in any message via `_URL_RE`, fetches up to 3 URLs concurrently with `_crawl4ai_fetch_async()` (async httpx). URL content is injected as a system context block into enriched history before routing, and into the system prompt for medium/complex agents.
- **Tier upgrade**: if URL content is successfully fetched, light tier is upgraded to medium (light model cannot process page content).
- **Complex agent tools**: `web_search` (SearXNG + Crawl4AI auto-fetch of top 2 results) and `fetch_url` (single-URL Crawl4AI fetch) remain available for the complex agent's agentic loop. Complex tier also receives the pre-fetched content in system prompt to avoid redundant re-fetching.
MCP tools from openmemory (`add_memory`, `search_memory`, `get_all_memories`) are **excluded** from agent tools — memory management is handled outside the agent loop.
### Fast Tools (`fast_tools.py`)
Pre-flight tools that run before the LLM in the `asyncio.gather` alongside URL fetch and memory retrieval. Each tool has a regex `matches()` classifier and an async `run()` that returns a context string injected into the system prompt. The router uses `FastToolRunner.any_matches()` to force medium tier when a tool matches.
| Tool | Trigger | Data source |
|------|---------|-------------|
| `WeatherTool` | weather/forecast/temperature keywords | SearXNG query `"погода Балашиха сейчас"` — Russian sources return °C |
| `CommuteTool` | commute/traffic/arrival time keywords | `routecheck:8090/api/route` — Yandex Routing API, Balashikha→Moscow center |
To add a new fast tool: subclass `FastTool` in `fast_tools.py`, add an instance to `_fast_tool_runner` in `agent.py`.
### `routecheck` service (`routecheck/app.py`)
Local web service that exposes Yandex Routing API behind an image captcha. Two access paths:
- **Web UI** (`localhost:8090`): solve PIL-generated arithmetic captcha → query any two lat/lon points
- **Internal API**: `GET /api/route?from=lat,lon&to=lat,lon&token=ROUTECHECK_TOKEN` — bypasses captcha, used by `CommuteTool`
Requires `.env`: `YANDEX_ROUTING_KEY` (free tier from `developer.tech.yandex.ru`) and `ROUTECHECK_TOKEN`. The container routes Yandex API calls through the host HTTPS proxy (`host.docker.internal:56928`).
### Medium vs Complex agent
| Agent | Builder | Speed | Use case |
|-------|---------|-------|----------|
| medium | `_DirectModel` (single LLM call, no tools) | ~3s | General questions, conversation |
| complex | `create_deep_agent` (deepagents) | Slow — multi-step planner | Deep research via `/think` prefix |
### Key files
- `agent.py` — FastAPI app, lifespan wiring, `run_agent_task()`, Crawl4AI pre-fetch, fast tools, memory pipeline, all endpoints
- `fast_tools.py` — `FastTool` base class, `FastToolRunner`, `WeatherTool`, `CommuteTool`
- `routecheck/app.py` — captcha UI + `/api/route` Yandex proxy
- `bifrost-config.json` — Bifrost provider config (Ollama GPU, retries, timeouts)
- `channels.py` — channel registry and `deliver()` dispatcher
- `router.py` — `Router` class: regex + LLM classification, light-tier reply generation
- `vram_manager.py` — `VRAMManager`: flush/poll/prewarm Ollama VRAM directly
- `agent_factory.py` — `build_medium_agent` (`_DirectModel`, single call) / `build_complex_agent` (`create_deep_agent`)
- `openmemory/server.py` — FastMCP + mem0 config with custom extraction/dedup prompts
- `wiki_research.py` — batch research pipeline using `/message` + SSE polling
- `grammy/bot.mjs` — Telegram long-poll + HTTP `/send` endpoint
@ARCHITECTURE.md

26
openmemory/CLAUDE.md Normal file
View File

@@ -0,0 +1,26 @@
# openmemory
FastMCP server wrapping mem0 for persistent per-session memory, backed by Qdrant + nomic-embed-text.
## Tools exposed (MCP)
- `add_memory(text, user_id)` — extract facts from a conversation turn and store in Qdrant
- `search_memory(query, user_id)` — semantic search, returns results with score ≥ 0.5
- `get_all_memories(user_id)` — dump all stored memories for a session
These are called directly by `agent.py` (outside the agent loop), never exposed to the LLM as tools.
## Two Ollama instances
- **GPU** (`OLLAMA_GPU_URL`, port 11436) — extraction model (`qwen2.5:1.5b`): pulls facts from conversation text
- **CPU** (`OLLAMA_CPU_URL`, port 11435) — embedding model (`nomic-embed-text`): 50–150 ms per query
## Prompts
Custom `EXTRACTION_PROMPT` starts with `/no_think` to suppress qwen3 chain-of-thought and get clean JSON output. Custom `UPDATE_MEMORY_PROMPT` handles deduplication — mem0 merges new facts with existing ones rather than creating duplicates.
## Notes
- Qdrant collection is created automatically on first use
- Memory is keyed by `user_id` which equals `session_id` in `agent.py`
- Extraction runs after the reply is sent (background task) — GPU contention with medium model is avoided since the semaphore is released before `_store_memory()` is scheduled

25
routecheck/CLAUDE.md Normal file
View File

@@ -0,0 +1,25 @@
# routecheck
FastAPI service providing a Yandex Routing API proxy behind an image captcha.
## Purpose
Yandex Routing API free tier requires a website that uses the API. This service is that website.
It also exposes an internal endpoint (`/api/route`) used by `CommuteTool` in `fast_tools.py`.
## Two access paths
- **Web UI** (`/`): solve PIL arithmetic captcha → get a token → query any two lat/lon points
- **Internal API**: `GET /api/route?from=lat,lon&to=lat,lon&token=$ROUTECHECK_TOKEN` — no captcha
## Key env vars
- `YANDEX_ROUTING_KEY` — from developer.tech.yandex.ru, Router API, free tier
- `INTERNAL_TOKEN` — equals `ROUTECHECK_TOKEN` from root `.env`; shared with deepagents
- `HTTPS_PROXY` — set to `http://host.docker.internal:56928`; container has no direct external internet
## Notes
- Captchas expire after 5 min, route tokens after 1 hour, both stored in-memory (restart clears them)
- Yandex API expects `lon,lat` order (not `lat,lon`) — `app.py` swaps automatically
- Captcha image endpoint: `GET /captcha/image/{id}` — regenerates on each call with random noise