navi: a token count that matches the model, a type for every file, and a socket that admits it is dead ...

Compression decided everything in chars // 3, and on this workload that sits
~1.8x below reality — 101k real prompt tokens against a ~56k estimate — so the
trigger, the target, the truncate budgets and the pre-call size check all ran
in a unit half the size of the real one. TokenCalibration now fits
overhead + chars / chars_per_token per model from the prompt_eval_count that
every call already reports, and a profile declares its own context_window
(sent to Ollama as num_ctx) instead of every model being held to OLLAMA_NUM_CTX.

Superseded summaries were never demoted in the database: save() only updates
flags for rows in session.messages, a summary is is_display=False, so it lived
in session.context alone and every reload resurrected it — prod sessions
carried up to 15 live summaries. PgSessionStore.set_context_flag() writes the
flip for those rows directly.

Five extension tables disagreed about what a file is. navi/content_types.py is
the one detector — signature, then extension, then a text probe — it decides
the MIME the download route serves, and the session file list now carries the
type the viewers branch on.

The idle receive loop answers client pings and sends its own heartbeat. A
half-open socket keeps readyState at OPEN and never fires close, so traffic
coming back is the only proof the connection still works.

compressor: single-summary invariant — old summaries always fold into the fresh one ...

partition_messages now strips is_summary messages out of the turn structure
before grouping: a summary is already-compressed history, not a user request.
Previously its role=user made it a turn head, so midturn compression kept the
old summary verbatim beside the fresh one (head kept + new summary added),
and adaptive swap resurrected summary turns whose text pattern-matches the
importance heuristics. This piled stale summary copies into the context
(one session accumulated 6 copies = 120k of 239k chars) and made compression
no-op for whole stretches before a lucky pass finally shrank it.

Also folds summaries on the early-exit path so two stale summaries
consolidate via the meta-summary, and fixes the demotion loop in
compress_and_save_session to cover session.context too: after a reload,
context-only rows (is_display=False summaries) were absent from
session.messages and never demoted, so is_context=True persisted in the DB
and the copies survived into the loaded context.

compress: route every trigger path through compress_and_save_session ...

- CompressionWorker keeps the real-token gate (ctx.context_tokens) but now
  delegates to compress_and_save_session(reason="postturn") — it used to
  call compress_context directly with its own save logic and silently
  no-op'ed on any summarizer failure, leaving the session over the
  threshold until the next turn's gates
- pre-turn and /compact gates use real_baseline_estimate instead of the
  chars//3 heuristic (mid-turn already did), so code-heavy contexts
  compress in time instead of tripping ContextTooLargeError
- summarizer input keeps head + tail instead of head-only truncation,
  which silently dropped the newest summarized messages closest to the
  keep window
- critical tool results over the 4000-char budget keep head+tail halves
  instead of collapsing to the 800-char non-critical preview (a 3999-char
  read survived whole, a 4001-char one lost 98%)
- images count 500 tokens each, not 500 per message carrying them
- _plan_compression is the single partition decision shared by
  compress_context and would_compress, so the dry-run prediction cannot
  drift from the real path; the trigger reason is now logged
  (preturn/midturn/postturn/forced)

compress: raise budget 6000 + richer prompt detail + input headroom ...

Increase the compression summary budget and push the prompt toward more
detail and completeness, so the agent keeps more durable facts when old turns
are folded into a summary.

Budget (output):
- context_summary_max_tokens 4000 → 6000 (global default).
- navi_code compression_max_tokens 4000 → 6000 (profile override).

Prompt (base, all profiles):
- New 'Intermediate Findings' section: durable facts from read/grep/log/terminal
  that informed decisions (was forbidden by 'do not preserve intermediate
  reasoning'). Existing sections get richer instructions (Active Files: key
  symbol affected; Decisions: rationale + trigger; Completed: step + file +
  verification outcome; Errors: snippet + what was tried + outcome).
- Loosen 'tight bullet points' → 'specific and complete; prefer listing over
  generalizing — losing a fact costs more than a longer summary'.
- 'Do not write implementation code' → 'preserve short critical code verbatim
  (final signatures, one-line fixes, key config lines); drop long blocks'.

Input headroom (so the summarizer can SEE what it must preserve):
- Non-critical tool result preview 300 → 800 chars.
- _MAX_SUMMARY_INPUT_CHARS 24000 → 32000 (fits 65k–128k windows with 6k output).

Profile prompt (navi_code/compression_prompt.txt): preserve short diagnostic
snippets verbatim (exact error line, failing assertion, one-line fix) instead
of blanket 'do not preserve long output'; keep code signatures + short key
snippets, not just names.

Tests: Intermediate Findings section in base prompt; 'short critical code'
phrasing; non-critical preview keeps 800 chars; meta-summary threshold raised
(2×6000 chars). Fixed flaky test_attach_session_seeds_context_fill (relied on
_startup worker timing after Etap 4 lengthened attach; now explicit attach
like its sibling + mock get_todos to skip a 30s×2 real-network timeout).
Full suite: 1008 passed, 1 skipped.

compressor: target hysteresis — shrink to 65% after compression, not just below the trigger ...

Compression triggered at 90% but left the context just under the trigger, so a
fixed keep_recent (navi_code: 12 turns ≈ 104 messages) re-triggered a couple
of messages later — the context yo-yoed at the trigger line instead of gaining
headroom. Add context_compression_target (0.65): in turn-based (preturn) mode
compress_context shrinks keep_recent until the verbatim kept region fits 65% of
the window, folding the extra turns into the same single summary LLM call; a
safety net in compress_session token-budget-truncates if the kept region alone
still exceeds the target (midturn kept a huge in-flight turn, or the
keep_recent floor can't fit). Trigger 90% → target 65% leaves real headroom.

compressor: split in-flight turn in midturn mode + flush incrementally ...

Two related fixes for long autonomous turns in planning profiles (navi_code):

A1 — partition_messages turn-based branch held the in-flight (current) turn
verbatim when turns > keep_recent, so midturn compression was shallow
(e.g. 150 -> 141 messages). Now, when keep_recent_messages is set, the
in-flight turn is split like partition_current_turn_messages: head (user
request) + tail (recent tool steps) kept, middle summarized. The adaptive
swap is disabled in midturn mode (it could move the in-flight turn into
old_turns and summarize the current request whole). First branch
(len(turns) <= keep_recent) is untouched, so locked-in midturn tests hold.

B1 — agent.run_stream had no try/finally around the for-loop, so an
asyncio.CancelledError (server restart/shutdown) unwound the stack with no
flush: all in-memory turn messages (sequence_number < 0) were lost, only the
user message survived. Add incremental save() after planning, after the
assistant tool-call decision, and after each tool result, so a crash loses at
most the single in-flight tool call, not the whole turn. B2 (try/except
safety-net) was dropped: B1 leaves no window where an append is unsaved.

Tests: midturn split with many turns (partition + compress_session), and
crash/cancel persistence via a snapshot session store that mirrors the DB
boundary.

Co-Authored-By: Claude <noreply@anthropic.com>

compression: profile-aware worker + real-token baseline estimator ...

Item 2 — thread the active profile into CompressionWorker so
compress_context applies per-profile overrides (compression_keep_recent,
compression_max_tokens, compression_prompt_file). navi_code now compresses
with keep_recent=12 instead of the global 8.

Item 1 — estimate the next LLM call's context from the *real* prompt_tokens
of the previous call (bulk) plus a heuristic delta for messages appended
since, replacing the chars//3 estimate that undercounts code-heavy tool
output and fired midturn compression too late (Navi kept working until the
window was exhausted). Baseline is recorded after each stream and cleared
after compression; check_context_size and the midturn gate use it, with a
heuristic fallback when no baseline exists or the context shrank.

Co-Authored-By: Claude <noreply@anthropic.com>

compression: fix auto-compress no-op on few-huge-messages + honest status ...

Root cause: the compression gate (should_compress) measures tokens, but the
partition measures message/turn count, and CompressionStarted was emitted
before the attempt. For navi_code's "few very large messages" shape (one big
file read = 1 user + assistant + 1 huge tool result, 66k tokens in 3 messages)
the gate fired, the UI showed "compression", but partition returned
to_summarize=[] -> compress_context None -> nothing shrank. The agent kept
going until the window overflowed. It wasn't running "during" compression —
there was no compression, just a no-op the user mistook for one.

A. Per-message head/tail truncation in context_builder.build(): oversized
   tool/assistant messages (over context_message_token_budget, 0=num_ctx//6)
   are capped head+marker+tail in the LLM view only (model_copy — stored
   history and reloads are never affected). A single huge tool result can no
   longer alone blow the window; user/system messages are never truncated.

B. Token-budget hard-truncate fallback in compress_session: when partition
   no-ops but tokens exceed the threshold, drop oldest turns to num_ctx*0.5.
   _hard_truncate is now token-aware (was a fixed message-count floor that
   no-oped on <=6 messages even when huge). New would_compress() predicts
   compress_session's real outcome with no LLM call.

C. Honest CompressionStarted: _compression_events_midturn/_preturn emit it
   only after would_compress() confirms the partition (or token-budget
   fallback) can actually shrink the stored context — no more "compression"
   status with no ContextCompressed to follow.

Bonus: post-turn CompressionWorker now passes keep_recent_messages=
max(12, context_keep_recent*2), matching the midturn path, so a single long
autonomous turn compresses post-turn too (was always a no-op).

Tests (+14): would_compress agreement, token-budget fallback, token-aware
hard_truncate, build() truncation (preserves user, no mutation, head+tail),
agent no-CompressionStarted-when-nothing-to-compress, worker single-long-turn.

Co-Authored-By: Claude <noreply@anthropic.com>

compressor: structured summaries, profile-aware compression, adaptive keep_recent ...

- Replace free-form summary with strict Markdown template (Goal, Active Files,
  Decisions, Completed Work, Pending Work/Todo, Errors, Key Values).
- Keep filesystem/code_exec/terminal tool results and messages with
  is_compression_critical=True verbatim during compression instead of 300-char truncation.
- Make compression profile-aware: AgentProfile gains compression_keep_recent,
  compression_max_tokens, compression_prompt_file. navi_code uses dedicated
  compression prompt and larger keep_recent/max_tokens.
- Adaptive partition_messages(): important turns (user corrections, errors,
  critical tools) survive longer; filler/social turns compress sooner.
- Increase default context_summary_max_tokens from 3000 to 4000.
- Propagate active profile changes to ContextCompressor and SubAgentRunner.

Co-Authored-By: Claude <noreply@anthropic.com>

agent: skip planning for casual greetings + strengthen DIRECT shortcut in Phase 1 ...

- Add fast _is_casual_message heuristic in navi/core/agent.py. Greetings and
  social chat (e.g. 'привет', 'как дела', 'hi', 'thanks') bypass planning even
  on the first session message, unless planning_mandatory is enabled.
- Strengthen Phase 1 planning prompt in navi/core/planning.py: explicitly
  require DIRECT output for greetings, simple questions, and one-step
  instructions.
- Fix broken ContextCompressed import in navi/core/compressor.py.
- Add unit tests covering the new heuristic and DIRECT prompt.

Co-Authored-By: Claude <noreply@anthropic.com>

Navi Code TUI: fix input box layout, command palette duplicate IDs, status renderer, and WS input loop ...

- clients/terminal/tui/widgets/input_box.py: switch Horizontal to Vertical with width: 100% for Input so it renders and accepts input in real terminals; add refresh on Input.Changed.
- clients/terminal/tui/screens/command_palette.py: remove fixed ListItem IDs to avoid DuplicateIds on fast filter.
- clients/terminal/tui/chat_model.py + renderers/status.py + widgets/chat_panel.py: render backend status events as dim system messages instead of raw dicts.
- clients/terminal/tui/ws_bridge.py: start NaviWebSocketClient.input_loop so enqueued user messages are actually sent to the backend.
- clients/terminal/tui/tui_app.py: focus InputBox synchronously in on_mount so typing works immediately.
- tests/clients/test_tui_app.py: regression test for visible input text.

Co-Authored-By: Claude <noreply@anthropic.com>

Fix 19 issues found in full codebase review ...

Backend:
- Stop session auth bypass: require auth for owned sessions, reject anonymous with 401
- upload_file: stream chunks directly to disk instead of buffering in RAM
- MCP config: validate name against path traversal regex
- auth deps: cleanup stale refresh locks periodically
- auth routes: expire mobile auth states after 10 min to prevent unbounded growth
- compressor: meta-summarize existing summaries before compression; preserve assistant content when tool_calls present; rewrite hard_truncate to keep whole turns
- orchestrator: configurable WS replay buffer size; async cleanup/remove_websocket/clear_busy; fix run_recall ContextVar order to avoid deadlock on _build_agent failure; await cleanup in finally
- agent: persist image_msg in session.messages; remove archived messages from session after archive; remove duplicate StreamStopped yield on tool stop
- websocket: try/except around create_task with cleanup on failure; await remove_websocket

Frontend:
- App.vue: hashchange listener lifecycle in onMounted/onUnmounted
- MessageList.vue: passive scroll, flash timeout cleanup, archive scroll snapshot
- InputBar.vue: 300 ms debounce on draft save to localStorage
- SessionList.vue: remove :key from DynamicScroller to avoid remount jitter

Tests: 422 passed, 1 skipped

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Add meta-summary for multi-level compression ...

When to_summarize contains multiple existing summary messages whose
combined length exceeds 8000 chars (~1/3 of max summarizer input),
run a quick meta-summary pass first to consolidate them into a single
compact summary before the main compression. This prevents information
loss when repeated compressions stack up long summary chains.

- _meta_summarize(): fast LLM pass (think=False, max_tokens=1500)
- compress_context(): detects >1 long summaries and triggers meta pass
- Graceful fallback: if meta-summary fails, continue with raw summaries
- 3 new unit tests: consolidation, skipped for short summaries, failure fallback

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Review fixes: restore _build_sessions, fix flags, search filter, tests ...

- Restored _load_messages_map and _build_sessions helpers that were
  accidentally dropped in Phase 5 (list_all/list_page/search_list
  called them but they were missing, causing NameError at runtime)
- _build_sessions now filters messages by is_display so list methods
  return consistent display-only history like get()
- count_all/search_list EXISTS subquery now filters to is_display=true
  so search only matches visible chat messages
- Updated pg_session_store docstring to remove stale dual-write claim
- compressor summary_msg now defaults to is_display=False
- Added unit tests for message flags (agent, compressor, planning)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Step 4: Extract SubAgentRunner from run_ephemeral() ...

- Create navi/core/subagent_runner.py with full sub-agent loop logic
- Move _iter_stream_guarded to navi/core/stream_guard.py
- Move _check_context_size to ContextCompressor.check_context_size()
- Extract build_tool_list() and load_user_enabled_tools() to tool_utils.py
- Agent.run_ephemeral() becomes a thin wrapper delegating to SubAgentRunner
- Remove ~310 lines from agent.py
- All existing run_ephemeral tests pass unchanged

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Extract ContextCompressor, fix STL viewer, expand test suite, add architecture audit docs ...

- Extract ContextCompressor from agent.py (Step 1 of god-object refactor)
- Add retry + hard-truncate fallback logic to ContextCompressor
- Add unit tests: agent loop (14), compressor (18), KV store (8),
  auth encrypt (3), auth deps (13), todo/scratchpad/image_view/memory
- Fix WebGL STL viewer: allow-same-origin sandbox + graceful fallback
- Add CompressionStarted event and client-side compression notice
- Add docs/architecture_weak_spots.md and plan_01_god_object_agent.md
- Update test_events.py and test_agent_context_size.py for new logic

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Fix ollama_backends / FallbackOllamaBackend issues ...

- registry.py: always use FallbackOllamaBackend (unified backend).
  Enables model priority lists in all deployments, not just multi-server.
- agent.py: add missing think=profile.think_enabled to run() (REST endpoint).
- compressor.py: fix model param type (str → list[str] | str | None).
- fallback.py: harden load_servers_from_file against missing/bad JSON files
  and entries without host. Add clear_blacklists() for manual reset.
- admin.py: add POST /admin/ollama/clear-blacklists endpoint.
- tech_debt_review: document dead stream() methods.
- tests: add tests for single-server fallback, bad file handling,
  missing host skipping, and blacklist clearing.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Refine 3D modeler workflow

Improve compression and memory prompts

Autonomous reasoning improvements: budget, anchoring, anti-stall, validation ...

- AgentProfile: per-profile thinking mechanics flags (think_enabled,
  iteration_budget_enabled, goal_anchoring, anti_stall, step_validation,
  planning_reflect, adaptive_replan) — all profiles updated in config.json
- Iteration budget: inject remaining iterations into context so model knows
  when to wrap up; urgency levels at ≤7 and ≤3 remaining
- Goal anchoring: inject original goal + todo state every N iterations to
  prevent drift on long tasks
- Anti-stall: two signals — no todo progress for N iterations, or identical
  tool calls repeated N times; warning injected into context
- Todo step validation: marking done requires a validation field describing
  how result was verified; failed gets a soft nudge with tip for re-planning
- stream_complete: add think param to base class, ollama and openai backends
- Summarizer: raise max_tokens 1024→3000, expand system prompt with
  user-preferences section and verbatim-value instructions
- Compression card: persist to session.messages (is_compression flag on
  Message), show expandable summary in webclient with markdown body
- ToolResult.to_message_content: always include output on failure so
  tracebacks and error details reach the model (fixes silent Error: None)
- Developer profile: fix subagent profile secretary→developer, add write_tool
  to subagent_tools, clarify write_tool vs filesystem in system prompt

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Planning phases, context compression, and tool improvements ...

Agent:
- Planning now a 3-phase async generator: Analysis → Execution plan → AIHelper critic
- Yield PlanningStatus events before each phase (UI progress labels)
- Phase 1 runs with think=True for deeper analysis
- Phase 2 includes available tool list so executor assignments are accurate
- Phase 3: independent critic pass validates and corrects TOOL: names against real tool list
- Planning converted from list return to async generator (fixes token accounting)

Backend:
- Context compression threshold: 80% → 70% to trigger earlier
- Compressor summary prompt: structured sections (goal, work state, key facts, outputs, errors)
- Terminal output capped at 5000 chars to prevent context flooding
- Web search: region=wt-wt for DDG, country=ALL for Brave, language=all for SearxNG
- Scratchpad: mandate writing a 'goal' section at start of multi-step tasks
- secretary max_iterations: 40→25, temperature: 0.7→0.5
- server_admin max_iterations: 40→20

Webclient:
- ThinkingCard strips <thought> XML tags leaked by Ollama
- planning_status WS event wired to chat.onPlanningStatus()

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Add explicit output token budget for summarizer (context_summary_max_tokens) ...

Previously there was no num_predict set for the summarization LLM call,
so Ollama used its server default (often 128 tokens — very short summaries).

- Add max_tokens param to LLMBackend.complete() and OllamaBackend (→ num_predict)
- Add context_summary_max_tokens: int = 1024 to config
- Thread it through compress_context() and CompressionWorker

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Expand summarization budget for better context quality ...

- _MAX_SUMMARY_INPUT_CHARS: 12k → 24k chars (2x input fed to summarizer)
- context_keep_recent: 10 → 8 turns (2 more turns go into each summary batch)
- Summarizer prompt: replace "Be brief" with "Be thorough" — capture code/config
  snippets and enough detail to continue the conversation without original messages

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Expose compression summary as collapsible debug card in chat UI ...

ContextCompressed event now carries the full summary text produced by the
LLM. Compression notice in chat becomes a <details> element showing
message count (before→after) with the summary expandable on click.
Rendered as markdown via marked.js.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Add stop button and fix context compression hang ...

Stop generation:
- Client: send button toggles to red ■ during streaming; sends {type:stop} via WS
- Server: _stream_recv concurrently reads incoming messages during streaming using
  asyncio.wait — stop signal is handled immediately without polling
- Cooperative stop via asyncio.Event (current_stop_event ContextVar): agent breaks
  out of LLM async-for cleanly so aclose() fires → Ollama stream closes gracefully,
  model stays in VRAM. No task.cancel() which would eject the model.
- StreamStopped event propagates through run_stream/run_ephemeral; sub-agents stop
  via the same shared stop_event inherited through task context

Context compression fix:
- compress_context passes think=False to llm.complete() — no extended reasoning
  during summarization which caused GPU hang
- Input truncated to 12k chars before sending to summarizer
- LLMBackend.complete() / OllamaBackend.complete() accept think: bool | None override

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Review fixes: events module, circular imports, deps, vision-aware compression ...

- Extract all AgentEvent dataclasses to navi/core/events.py; import from
  there in agent.py and __init__.py — eliminates circular import between
  workers and core
- workers/compressor.py: remove runtime import hack, use navi.core.events
- workers/base.py: WorkerResult.events typed as list[AgentEvent] (was Any)
- api/deps.py: replace @lru_cache on mutable list with module-level
  singletons (_registries, _workers)
- core/compressor.py: _format_for_summary returns (text, images); images
  passed to summarization LLM so vision models describe them in summary;
  non-vision models silently ignore the images field; docstring updated
- client/js/app.js: add comment explaining is_summary backward compat branch

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Separate display history from LLM context; formalize worker system ...

Architecture change:
- session.messages: full display history, never modified by compression
- session.context: what the LLM sees, may be compressed by workers
- System messages go only into context (not display history)
- Image injections (synthetic) go only into context
- User/assistant/tool messages go into both

SQLite: add context column with backward-compat migration
(empty context → initialized from messages on load)

Workers (navi/workers/):
- Worker ABC + WorkerContext + WorkerResult (base.py)
- CompressionWorker: compresses session.context when above threshold
- build_default_workers() returns [CompressionWorker()]
- Agent accepts workers list, runs them after StreamEnd
- Workers injected via deps.py get_workers() (lru_cached singleton)
- WebSocket agent construction also receives workers

Compressor: compress_context() now takes context[], not messages[]
Config: context_keep_recent 6 → 10
Agent: _run_workers() collects events from all workers and yields them

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Add context compression: rolling summarization when context fills up ...

Mechanism:
- After streaming ends, if context_tokens >= threshold (80% of num_ctx),
  compress old turns into a summary message using the same LLM
- Partition: keep system msg + last N turns verbatim (default 6);
  everything older goes to the summarizer
- Tool call groups (assistant + tool results) never split across boundary
- Existing summary messages folded into new compression pass — no stack growth
- Summary stored as Message(role=user, is_summary=True) after system msg
- On failure: logged, session left unchanged (non-fatal)

New files:
- navi/core/compressor.py: should_compress, partition_messages,
  compress_session (pure logic, testable without agent)

New config (navi/config.py):
- context_compression_enabled: bool = True
- context_compression_threshold: float = 0.80
- context_keep_recent: int = 6
- context_summary_temperature: float = 0.3

New agent event: ContextCompressed(messages_before, messages_after)
Message.is_summary: bool field marks compressed history blocks

Client:
- context_compressed WS event → subtle inline notice in message list
- loadHistory: is_summary messages rendered as collapsible summary cards
- style.css: .summary-card, .compression-notice

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>